Skip to content

armgit / AiSOC

Public Imported from GitHub
0

Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.

feat(hunts): grow the corpus to 68 and make its negatives mean something (#930)

Beenu Arora29acfe25a0fa
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index d06e336dc..4d6ba09c5 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -657,6 +657,18 @@ jobs:
run: |
python3 scripts/check_agent_read_tools.py --self-test
python3 scripts/check_agent_read_tools.py
+ # A hunt's negative scenario has to fail for the reason the hunt is
+ # about. A negative drawn from an unrelated log source never fires on
+ # anything, so a corpus of those scores a perfect false-positive rate
+ # while grading nothing. `run_evals.py --suite hunt_corpus` cannot see
+ # that, because it only asks whether the negative fired; this reads both
+ # scenarios and compares them field by field.
+ - name: Hunt negatives invert the clause their hunt is about
+ run: |
+ python3 -m pip install --quiet pyyaml pydantic
+ python3 scripts/check_hunt_scenarios.py --self-test
+ python3 scripts/check_hunt_scenarios.py
+ python3 scripts/run_hunt_evals.py --self-test
- name: Support bundle must not leak
run: |
python3 -m pip install --quiet pytest
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7dded6950..1200236eb 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,56 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
+### Added
+
+- **The hunt corpus grew from 5 hunts to 68, and the grading that covers it
+ stopped being satisfiable by accident** (gap-closure Phase 8.4).
+
+ **The corpus.** 63 new hunts across the ATT&CK tactics and the log sources
+ AiSOC ships connectors for: Windows Security and Sysmon, Linux auditd, AWS
+ CloudTrail, GCP audit, Kubernetes audit, Okta and Entra ID, Microsoft 365
+ and Google Workspace, Snowflake, Salesforce, Slack, Zoom, Confluence,
+ GitHub and GitLab audit, Vault, Duo, Jamf, Netskope, Tailscale, osquery,
+ DNS, proxy, firewall, VPN and the AI gateway. Every one ships a positive
+ and a negative synthetic scenario and is graded by
+ `services/agents/tests/test_hunt_corpus.py`.
+
+ **What 68 means, and what it does not.** Each hunt was replayed against its
+ scenario and observed to fire. That is not a claim any of them fires on a
+ given deployment's telemetry, which depends on whether that deployment's
+ connectors emit the fields the hunt names. `hunts/README.md` says so rather
+ than leaving the count to imply the stronger claim.
+
+ **The grading was satisfiable without testing anything.** A hunt passed when
+ it fired on its positive scenario and not on its negative, and the second
+ half is free: a negative drawn from a different log source never fires on
+ any hunt, so a corpus of those reports a perfect false-positive rate while
+ grading only the hunt's aim. This is the same circularity as the roughly
+ 600 detection fixtures synthesised from the rule they test.
+
+ `scripts/check_hunt_scenarios.py` closes it. It runs the production hunt
+ matcher over both scenarios and requires the negative to differ from its
+ positive in **exactly one indicator field**, that field not being the one
+ selecting the log source, with a near-miss floor beneath it. The rule is the
+ one the adversarial injection corpus already uses for its clean twins. It
+ found **8 violations on its first run against a corpus that was passing the
+ existing grading**, three of them in hunts that predate this phase: the DNS
+ tunnelling, OAuth mass-consent and anomalous role-assumption negatives each
+ differed from their positive in two clauses, so none of them said which
+ clause the hunt was actually discriminating on. All eight were rewritten to
+ invert one clause and hold the rest constant.
+
+ **`scripts/run_hunt_evals.py` now exists.** `hunts/README.md` had told
+ contributors to run it before opening a PR for months, and it had never been
+ written, so the one instruction a contributor was most likely to follow
+ failed at the shell. It is a thin wrapper over the `hunt_corpus` suite in
+ `scripts/run_evals.py` rather than a second scorer, for the reason
+ `scripts/run_model_matrix.py` is a wrapper: two scorers for one corpus drift,
+ and then two numbers describe the same thing and disagree. Its `--self-test`
+ parses its own source and asserts it imports neither the hunt engine nor the
+ loader, so it cannot score even if someone later wanted it to.
+
+
### Fixed
- **Five surfaces kept publishing 14 CORE services after ADR-0007 moved it to
diff --git a/GAP_CLOSURE_PROGRESS.md b/GAP_CLOSURE_PROGRESS.md
index ac43633b4..ecf50407d 100644
--- a/GAP_CLOSURE_PROGRESS.md
+++ b/GAP_CLOSURE_PROGRESS.md
@@ -426,7 +426,11 @@ suites' job, and anything at all against a live vendor MCP server.
- [ ] **8.1 Retro-hunts** consuming the `NEW_IOC` events nothing consumes today, with provenance, dedup, rate limits and budgets.
- [ ] **8.2 KEV exposure** checked against asset and vulnerability data, opening a case task when exposed.
- [ ] **8.3 Hunting agent** turning a hypothesis into a plan, read-only queries and evidence-backed findings; new `aisoc-hunt` role alias.
-- [ ] **8.4 Hunt library** grown from 5 to at least 50, each with positive and negative synthetic scenarios, and `hunts/README.md` corrected where it cites a script that does not exist.
+- [x] **8.4 Hunt library** grown from 5 to at least 50, each with positive and negative synthetic scenarios, and `hunts/README.md` corrected where it cites a script that does not exist.
+ - **68 hunts**, 63 new, across the ATT&CK tactics and the log sources AiSOC ships connectors for. Executable means replayed against a scenario and observed to fire; it is not a claim a hunt fires on a given deployment's telemetry, and `hunts/README.md` states the distinction.
+ - `scripts/check_hunt_scenarios.py` is the anti-tautology gate: a negative must differ from its positive in **exactly one indicator field**, and that field must not be the log-source selector. The existing grading could not see this, because a negative from an unrelated log source never fires on anything.
+ - The gate found **8 violations on its first run** against a corpus that was passing the existing grading, **3 of them in the 5 hunts that predate this phase**. All were rewritten to invert one clause.
+ - `scripts/run_hunt_evals.py` now exists (`hunts/README.md` had cited it for months). A thin wrapper over `run_evals.py --suite hunt_corpus`, with a `--self-test` that parses its own source and asserts it cannot score.
## Phase 9: Detection-engineering loop
diff --git a/ROADMAP.md b/ROADMAP.md
index 6be08735b..58871cea5 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -44,7 +44,7 @@ A proof-first, security-first program to make every README claim gate-backed, cl
- **Phase D (breadth):** D1 eight connectors (QRadar/Exabeam/Securonix/Devo/Netskope/Windows-Sysmon/Zeek-Suricata/syslog-CEF) · D2 AI/LLM-usage audit connector + `llm-*` detections + hot/cold lake tiering · D3 live-vendor mock-server smoke.
- **Phase E (prove it):** E1 CI-gated benchmark scoreboard tied to a deterministic live-agent MITRE-accuracy run.
-The claim-to-gate matrix stands at **216 rows — 208 GATED / 8 PARTIAL / 0 NO GATE** — **every product claim is backed by a failing test** and the ratchet (`MAX_NO_GATE=0`) forbids any regression. The 8 remaining PARTIAL rows are honest, named deferrals; none was relabelled without building the gate it names. Count the table rows with `python3 scripts/check_claim_gate_matrix.py` rather than trusting a figure quoted in prose — this line has gone stale before. The six lettered deferrals are scoped in [`docs/audit/DEFERRED_SUBPHASES.md`](docs/audit/DEFERRED_SUBPHASES.md).
+The claim-to-gate matrix stands at **219 rows — 211 GATED / 8 PARTIAL / 0 NO GATE** — **every product claim is backed by a failing test** and the ratchet (`MAX_NO_GATE=0`) forbids any regression. The 8 remaining PARTIAL rows are honest, named deferrals; none was relabelled without building the gate it names. Count the table rows with `python3 scripts/check_claim_gate_matrix.py` rather than trusting a figure quoted in prose — this line has gone stale before. The six lettered deferrals are scoped in [`docs/audit/DEFERRED_SUBPHASES.md`](docs/audit/DEFERRED_SUBPHASES.md).
## v4.0 — Shipped
diff --git a/apps/docs/docs/compliance/evidence-pack.md b/apps/docs/docs/compliance/evidence-pack.md
index 832a493d2..bf37af61d 100644
--- a/apps/docs/docs/compliance/evidence-pack.md
+++ b/apps/docs/docs/compliance/evidence-pack.md
@@ -33,7 +33,7 @@ and says so.
| Prompt-injection resistance | Three adversarial corpora: 32 payloads scanned in isolation and 71 incident pairs whose payloads are placed in attacker-controllable fields, both against a clean twin with a recall floor and a false-positive ceiling, plus 34 held-out pairs authored after the guard was frozen and carrying no floor at all, on which the guard measures 7.1% | `services/agents/tests/adversarial/` |
| Claim-to-gate traceability | Every product claim mapped to the CI job that fails when it stops being true | `docs/audit/CLAIM_TO_GATE_MATRIX.md` |
-The last row is the one worth reading first. It is the honest index: 208 rows
+The last row is the one worth reading first. It is the honest index: 211 rows
`GATED`, 8 `PARTIAL` with the specific gap named, and a ratchet that refuses
to let a row sit at `NO GATE`.
diff --git a/docs/audit/CLAIM_TO_GATE_MATRIX.md b/docs/audit/CLAIM_TO_GATE_MATRIX.md
index 18ed28521..d32c8d98e 100644
--- a/docs/audit/CLAIM_TO_GATE_MATRIX.md
+++ b/docs/audit/CLAIM_TO_GATE_MATRIX.md
@@ -224,6 +224,9 @@ Statuses: `GATED` (a CI job fails when the claim stops being true) · `PARTIAL`
| Auto-triage reads the tenant's knowledge-base runbooks and cites the chunks it was given | `apps/docs/docs/console/triage-context.md` (Knowledge-base runbooks) | `ci.yml` agents job (`test_triage_knowledge_base.py`: the block reaches the prompt ahead of the telemetry, every chunk carries a marker resolving to a document id **and chunk index**, the prompt names the markers the model may cite, and a marker no retrieved chunk carries is reported on the verdict's basis rather than left reading like one that resolves; an empty retrieval renders nothing at all, because "Runbooks: none" is a claim about the tenant) + `ci.yml` api job (`test_kb_triage_retrieval_route.py`: service-token only, a valid console session refused, tenant and query bound rather than interpolated) | GATED | The agents service had never read the knowledge base. Retrieval is the API's own full-text query and ranking rather than a second implementation, so a chunk triage was given is one an analyst searching by hand would have found |
| Retrieved knowledge-base text is contained as untrusted evidence, not as first-party guidance | `apps/docs/docs/console/triage-context.md` (How runbook text is contained) | `ci.yml` agents job (`test_triage_knowledge_base.py::TestContainment`: the block is nonce-fenced with the boundary sentence stated inline; a chunk echoing the fence is refused; **a leaked nonce is stripped even for a chunk the guard passes**, which is the property that matters because the guard's own held-out rate is 2 of 28; a high-severity chunk is dropped rather than demoting the alert, and the refusal is published on the verdict's basis. The scan runs on the raw text, pinned by a payload the *sanitiser* also recognises: sanitising first rewrites it to a redaction marker, the guard then reads a clean string, and the whole "ignore previous" family would score zero forever while the counter reported a clean library) | GATED | More containment than tenant-skill text gets, because a runbook is long, often imported in bulk, edited by more people, and reaches the prompt as prose that routinely quotes attacker output. Dropping the chunk rather than demoting the case is deliberate: demoting would let anyone who can write a runbook switch off auto-close for the tenant |
| Knowledge-base runbooks are point-in-time during a replay, and the freeze publishes what it refused | `apps/docs/docs/console/triage-context.md` (Freezing context for a replay); `services/agents/app/replay/shadow.py` | `ci.yml` agents job (`test_replay_leakage.py`: a runbook created after the split never reaches the prompt, with a **sensitivity half** running the live reader and asserting the same poison does; a reply that names no cutoff is counted as `cutoff_not_honoured` rather than as frozen, with a sensitivity half asserting an honoured cutoff does not raise it) + `integration.yml` (`tests/isolation/test_kb_retrieval_cutoff_live.py` runs the route's own `triage_retrieval_sql` against live Postgres: unscoped read first so an empty table cannot pass for isolation, then the cutoff, then the refusal count, and a cutoff before everything still returns its counts. Proven able to fail by deleting the cutoff predicate, by hardcoding the refusal count, and by neutralising the tenant predicate) + `ci.yml` gates job (`scripts/check_triage_context_freeze.py`: `CONTEXT_FREEZE_KINDS` classifies every source, a cutoff source's protocol method may not accept the instant from its caller, its frozen implementation must bind `as_of` to the snapshot's `split_at`, and a snapshot source's frozen implementation may await nothing. Twelve regressions injected into a detached copy, all twelve caught) | GATED | A runbook is usually written after the incident that prompted it, so this is the most natural way a replay could flatter itself. The refusal count is what makes the claim checkable: a cutoff that matched nothing and a cutoff that refused fifty documents return the same empty list |
+| The hunt corpus holds 68 hunts, every one graded in both directions | `hunts/README.md`; `plans/aisoc_gap_closure_plan.plan.md` (8.4) | `ci.yml :: p1-eval` (`run_evals.py --suite hunt_corpus` — every hunt must fire on its declared positive scenario and none may fire on its negative, with telemetry orphaned by a renamed hunt failing in the other direction) + `ci.yml :: python-lint` (`check_hunt_scenarios.py`) | GATED | 68 is hunts replayed against a scenario and observed to fire. It is **not** a claim any of them fires on a given deployment's telemetry, which depends on whether that deployment's connectors emit the fields the hunt names; `hunts/README.md` states the distinction rather than letting the count imply the stronger claim |
+| A hunt's negative scenario inverts the clause the hunt is about | `hunts/README.md` (Adding a hunt); `plans/aisoc_gap_closure_plan.plan.md` (8.4) | `ci.yml :: python-lint` (`check_hunt_scenarios.py` runs the **production** hunt matcher over both scenarios and requires the negative to differ from its positive in exactly one indicator field, that field not being the log-source selector, with a near-miss floor beneath it) | GATED | Proven against the pre-fix tree rather than only against the tree as committed: it found 8 violations on first run, 3 of them in hunts that predate this phase. The grading suite cannot see this — it asks only whether the negative fired, and a negative drawn from an unrelated log source never fires on anything, so a corpus of those reports a perfect false-positive rate while testing nothing |
+| `scripts/run_hunt_evals.py` exists and defines no scoring of its own | `hunts/README.md` | `ci.yml :: python-lint` (`run_hunt_evals.py --self-test` parses its own source with `ast` and asserts it imports the corpus grader and imports neither the hunt engine nor the loader, so it cannot score) | GATED | `hunts/README.md` cited this script for months and it had never existed, so the one instruction a contributor was most likely to follow failed at the shell. It is a wrapper for the same reason `run_model_matrix.py` is: a second scorer drifts from the one the scoreboard publishes |
## Summary
@@ -231,7 +234,7 @@ Statuses: `GATED` (a CI job fails when the claim stops being true) · `PARTIAL`
## Summary
-- GATED: 208
+- GATED: 211
- PARTIAL: 8
- NO GATE: 0 (**every claim is backed by a failing test.** The last NO GATE — the weekly benchmark scoreboard running live against `main` — closed in Phase E1: `scripts/check_scoreboard.py` ties the published scoreboard to a deterministic per-PR live-agent MITRE-accuracy run, while the funded weekly `wet-eval.yml` appends the LLM-tier rows. The remaining PARTIAL rows are honest, named deferrals — each states the specific gap and what closes it — not unproven claims.)
diff --git a/docs/audit/REPOSITORY_REALITY.md b/docs/audit/REPOSITORY_REALITY.md
index a2eb4d13b..67eb83ebe 100644
--- a/docs/audit/REPOSITORY_REALITY.md
+++ b/docs/audit/REPOSITORY_REALITY.md
@@ -125,6 +125,7 @@ SSRF allow-list entry, and is commented as such.
| Rules the engine loads | **2603** | **WORKING** — each was replayed through its real connector and the real engine and watched to fire |
| Detection rules on disk | **6991** | the rest are imports in languages with **no evaluator** here (Splunk SPL, Chronicle YARA-L, CAR pseudocode); the engine never reads them |
| YAML files under `detections/` | **7016** | the extra 25 are response playbooks, not rules — see `detections/playbooks/` |
+| Hunts under `hunts/` | **68** | **WORKING** — each replayed against a synthetic scenario and observed to fire, and each graded in the other direction too. Not a claim a hunt fires on a given deployment's telemetry, which depends on whether its connectors emit the fields the hunt names. The negative half of that grading was satisfiable without testing anything until Phase 8.4: a negative drawn from a different log source never fires on any hunt, so `check_hunt_scenarios.py` now requires it to differ from its positive in exactly one indicator field. It found 8 violations on first run, 3 in the 5 hunts that predate the phase |
The gap matters and has been published both ways in the past. `make stats`
reports the first two numbers side by side so they cannot be conflated again.
diff --git a/hunts/README.md b/hunts/README.md
index a0cf6e6d3..b003bc094 100644
--- a/hunts/README.md
+++ b/hunts/README.md
@@ -5,10 +5,17 @@ single threat hypothesis the agent layer can run on a schedule, scoring the
target telemetry corpus against a set of expected indicators.
The corpus is part of the same **eval-graded** pipeline as detections and
-prompts: every hunt ships with a synthetic positive scenario in
-`services/agents/tests/eval_data/synthetic_telemetry.jsonl` and is scored by
-`scripts/run_hunt_evals.py` (or as part of `scripts/run_evals.py` when run in
-`--all` mode).
+prompts. Every hunt ships with a synthetic positive **and** negative scenario
+in `services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl`, and both
+are graded by `scripts/run_hunt_evals.py` (or by
+`scripts/run_evals.py --suite hunt_corpus`, which owns the floors CI gates on;
+the hunt script is a thin wrapper over it rather than a second scorer).
+
+There are **68 hunts** at the time of writing. Every one of them is executable
+in the sense the rest of this repository uses the word: it was replayed
+against a scenario and observed to fire. That is not a claim any of them will
+fire on your telemetry, which depends on whether your connectors emit the
+fields the hunt names.
## Schema
@@ -86,11 +93,21 @@ On each fire the hunt agent:
## Adding a hunt
1. Drop a new YAML into `hunts/`.
-2. Add at least one synthetic positive scenario to
- `services/agents/tests/eval_data/synthetic_telemetry.jsonl` with the matching
- `incident_id` and `template_id`.
-3. Add a synthetic negative scenario (benign events that match *some* but not
- all indicators).
-4. Run `python scripts/run_hunt_evals.py` locally to confirm the hunt scores
- 100% true-positive and 0% false-positive.
-5. Open a PR — CI will re-run the hunt evals and gate on regression.
+2. Add a synthetic positive scenario to
+ `services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl` with the
+ matching `incident_id` and `template_id`. Every event there carries
+ `is_synthetic: true`, and the id must start with `INC-HUNT-`.
+3. Add a synthetic negative scenario, and **invert the clause the hunt is
+ about**. This is the part that is easy to get wrong in a way nothing
+ notices: a negative drawn from a different log source will never fire on
+ any hunt, so a corpus of those reports a perfect false-positive rate while
+ testing nothing. `scripts/check_hunt_scenarios.py` enforces the rule that
+ makes a negative meaningful — it must differ from its positive in
+ **exactly one indicator field**, and that field must not be the one
+ selecting the log source. A near miss is the only negative that grades a
+ hunt's discrimination rather than its aim.
+4. Run `python scripts/run_hunt_evals.py` to confirm the hunt fires on its
+ positive and stays quiet on its negative, then
+ `python scripts/check_hunt_scenarios.py` to confirm the negative is a near
+ miss rather than an unrelated event. The first cannot tell you the second.
+5. Open a PR. CI runs both.
diff --git a/hunts/ad-enumeration-burst.yaml b/hunts/ad-enumeration-burst.yaml
new file mode 100644
index 000000000..7620d50af
--- /dev/null
+++ b/hunts/ad-enumeration-burst.yaml
@@ -0,0 +1,41 @@
+id: hunt-ad-enumeration-burst
+name: Directory enumeration burst from one host
+description: |
+ Identity-management servers enumerate the directory by design. A marketing workstation reading three thousand objects in ten minutes is running a discovery tool.
+version: 1.0.0
+severity: medium
+category: identity
+tags:
+ - mitre.attack.t1087.002
+ - hunt.identity
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one host enumerate an unusual number of directory objects in a short window?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4661, 4662]
+ - field: "objectsEnumeratedLast10Min"
+ gte: 500
+ - field: "hostClass"
+ equals: "workstation"
+expected:
+ positive_incident_id: INC-HUNT-029
+ positive_template_id: hunt-ad-enumeration-burst
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-029-NEG
+ # The negative inverts hostClass, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1087/002/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/ai-gateway-prompt-with-credential-shape.yaml b/hunts/ai-gateway-prompt-with-credential-shape.yaml
new file mode 100644
index 000000000..911a39001
--- /dev/null
+++ b/hunts/ai-gateway-prompt-with-credential-shape.yaml
@@ -0,0 +1,39 @@
+id: hunt-ai-gateway-prompt-with-credential-shape
+name: Prompt submitted to a model containing a credential-shaped string
+description: |
+ Pasting a config file into a model to debug it is a common habit. Whether the model runs inside the estate decides whether that is a disclosure.
+version: 1.0.0
+severity: high
+category: application
+tags:
+ - mitre.attack.t1552.001
+ - hunt.ai
+log_sources:
+ - ai_gateway
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did anyone paste something credential-shaped into a model prompt?
+ indicators:
+ - field: "source"
+ equals: "ai_gateway"
+ - field: "promptContainsSecretShape"
+ equals: true
+ - field: "destinationIsSelfHosted"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-065
+ positive_template_id: hunt-ai-gateway-prompt-with-credential-shape
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-065-NEG
+ # The negative inverts destinationIsSelfHosted, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1552/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/ai-gateway-spend-spike.yaml b/hunts/ai-gateway-spend-spike.yaml
new file mode 100644
index 000000000..9870d1c5e
--- /dev/null
+++ b/hunts/ai-gateway-spend-spike.yaml
@@ -0,0 +1,39 @@
+id: hunt-ai-gateway-spend-spike
+name: Model spend for one identity far above its own baseline
+description: |
+ A busy pipeline makes many requests legitimately, so volume alone says little. Spend far above the identity's own baseline is what a leaked key looks like.
+version: 1.0.0
+severity: medium
+category: application
+tags:
+ - mitre.attack.t1496
+ - hunt.ai
+log_sources:
+ - ai_gateway
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one identity's model spend jump far above its own recent baseline?
+ indicators:
+ - field: "source"
+ equals: "ai_gateway"
+ - field: "spendRatioToBaseline"
+ gte: 20
+ - field: "requestsLastHour"
+ gte: 500
+expected:
+ positive_incident_id: INC-HUNT-066
+ positive_template_id: hunt-ai-gateway-spend-spike
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-066-NEG
+ # The negative inverts spendRatioToBaseline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1496/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/archive-created-then-uploaded.yaml b/hunts/archive-created-then-uploaded.yaml
new file mode 100644
index 000000000..cf3cbf16f
--- /dev/null
+++ b/hunts/archive-created-then-uploaded.yaml
@@ -0,0 +1,42 @@
+id: hunt-archive-created-then-uploaded
+name: Archive created locally and uploaded to a personal cloud service
+description: |
+ Large uploads to corporate storage are work. The same volume to a consumer service is the last step of most data-theft cases.
+version: 1.0.0
+severity: high
+category: data-exfil
+tags:
+ - mitre.attack.t1560.001
+ - hunt.data
+log_sources:
+ - sysmon
+ - proxy
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an archive created and then uploaded to a consumer file-sharing service?
+ indicators:
+ - field: "source"
+ equals: "proxy"
+ - field: "http_method"
+ equals: "POST"
+ - field: "destination_category"
+ equals: "personal-storage"
+ - field: "upload_bytes"
+ gte: 50000000
+expected:
+ positive_incident_id: INC-HUNT-039
+ positive_template_id: hunt-archive-created-then-uploaded
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-039-NEG
+ # The negative inverts destination_category, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1560/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/aws-role-trust-policy-widened.yaml b/hunts/aws-role-trust-policy-widened.yaml
new file mode 100644
index 000000000..768642027
--- /dev/null
+++ b/hunts/aws-role-trust-policy-widened.yaml
@@ -0,0 +1,39 @@
+id: hunt-aws-role-trust-policy-widened
+name: IAM role trust policy widened to an external principal
+description: |
+ Trust policies are edited as part of ordinary integration work. Trusting a principal outside the organisation hands standing access to somebody else's account.
+version: 1.0.0
+severity: critical
+category: cloud
+tags:
+ - mitre.attack.t1098.001
+ - hunt.cloud
+log_sources:
+ - aws_cloudtrail
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a role's trust policy changed to allow an account outside the organisation to assume it?
+ indicators:
+ - field: "source"
+ equals: "aws_cloudtrail"
+ - field: "eventName"
+ in: ["UpdateAssumeRolePolicy", "CreateRole"]
+ - field: "trustsExternalPrincipal"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-050
+ positive_template_id: hunt-aws-role-trust-policy-widened
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-050-NEG
+ # The negative inverts trustsExternalPrincipal, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1098/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/backup-job-disabled.yaml b/hunts/backup-job-disabled.yaml
new file mode 100644
index 000000000..f05745c77
--- /dev/null
+++ b/hunts/backup-job-disabled.yaml
@@ -0,0 +1,41 @@
+id: hunt-backup-job-disabled
+name: Scheduled backup job disabled or deleted
+description: |
+ Backup jobs are disabled for maintenance regularly. Outside a change window it is the step that makes encryption stick.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1490
+ - hunt.endpoint
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a backup job disabled outside a change window?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4701, 4699]
+ - field: "TaskName"
+ regex: "(?i)backup"
+ - field: "duringChangeWindow"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-048
+ positive_template_id: hunt-backup-job-disabled
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-048-NEG
+ # The negative inverts duringChangeWindow, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1490/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/beacon-regular-interval.yaml b/hunts/beacon-regular-interval.yaml
new file mode 100644
index 000000000..b9997b834
--- /dev/null
+++ b/hunts/beacon-regular-interval.yaml
@@ -0,0 +1,42 @@
+id: hunt-beacon-regular-interval
+name: Outbound connections at a suspiciously regular interval
+description: |
+ Telemetry agents beacon on a fixed timer too. Whether the destination is a known service is what separates monitoring from a command channel.
+version: 1.0.0
+severity: high
+category: network
+tags:
+ - mitre.attack.t1071.001
+ - hunt.network
+log_sources:
+ - proxy
+ - firewall
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did a host make outbound connections at near-constant intervals over hours?
+ indicators:
+ - field: "source"
+ equals: "proxy"
+ - field: "connection_count_last_6h"
+ gte: 100
+ - field: "interval_jitter_ratio"
+ lte: 0.05
+ - field: "destination_is_known_service"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-041
+ positive_template_id: hunt-beacon-regular-interval
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-041-NEG
+ # The negative inverts destination_is_known_service, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1071/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/cloud-access-key-created-then-used-elsewhere.yaml b/hunts/cloud-access-key-created-then-used-elsewhere.yaml
new file mode 100644
index 000000000..885846a9f
--- /dev/null
+++ b/hunts/cloud-access-key-created-then-used-elsewhere.yaml
@@ -0,0 +1,39 @@
+id: hunt-cloud-access-key-created-then-used-elsewhere
+name: New access key used from a different network than the one that created it
+description: |
+ Key creation is routine in any automated estate. A key whose first use comes from a different network than its creation is the shape of a stolen credential rather than a rotation.
+version: 1.0.0
+severity: high
+category: cloud
+tags:
+ - mitre.attack.t1098.001
+ - hunt.cloud
+log_sources:
+ - aws_cloudtrail
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a new access key created and then used from a different source address?
+ indicators:
+ - field: "source"
+ equals: "aws_cloudtrail"
+ - field: "eventName"
+ in: ["CreateAccessKey"]
+ - field: "firstUseFromDifferentAsn"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-010
+ positive_template_id: hunt-cloud-access-key-created-then-used-elsewhere
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-010-NEG
+ # The negative inverts firstUseFromDifferentAsn, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1098/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/cloud-permission-enumeration.yaml b/hunts/cloud-permission-enumeration.yaml
new file mode 100644
index 000000000..f8367e42d
--- /dev/null
+++ b/hunts/cloud-permission-enumeration.yaml
@@ -0,0 +1,41 @@
+id: hunt-cloud-permission-enumeration
+name: Identity enumerated its own permissions across many services
+description: |
+ Posture scanners sweep every service on a schedule and look identical to an attacker orienting. Which identity is doing it is the whole signal.
+version: 1.0.0
+severity: medium
+category: cloud
+tags:
+ - mitre.attack.t1069.003
+ - hunt.cloud
+log_sources:
+ - aws_cloudtrail
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one identity make a burst of read-only describe or list calls across unrelated services?
+ indicators:
+ - field: "source"
+ equals: "aws_cloudtrail"
+ - field: "eventName"
+ regex: "^(Describe|List|Get)"
+ - field: "distinctServicesLast15Min"
+ gte: 12
+ - field: "performedByPipeline"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-030
+ positive_template_id: hunt-cloud-permission-enumeration
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-030-NEG
+ # The negative inverts performedByPipeline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1069/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/cloudtrail-logging-stopped.yaml b/hunts/cloudtrail-logging-stopped.yaml
new file mode 100644
index 000000000..daca27f7e
--- /dev/null
+++ b/hunts/cloudtrail-logging-stopped.yaml
@@ -0,0 +1,39 @@
+id: hunt-cloudtrail-logging-stopped
+name: Cloud audit trail stopped or deleted
+description: |
+ Infrastructure pipelines reconfigure trails as part of ordinary change. A human identity stopping one is the first move in making the rest invisible.
+version: 1.0.0
+severity: critical
+category: cloud
+tags:
+ - mitre.attack.t1562.008
+ - hunt.cloud
+log_sources:
+ - aws_cloudtrail
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an audit trail stopped, deleted, or had its logging disabled?
+ indicators:
+ - field: "source"
+ equals: "aws_cloudtrail"
+ - field: "eventName"
+ in: ["StopLogging", "DeleteTrail", "PutEventSelectors"]
+ - field: "performedByPipeline"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-022
+ positive_template_id: hunt-cloudtrail-logging-stopped
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-022-NEG
+ # The negative inverts performedByPipeline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1562/008/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/confluence-space-permissions-opened.yaml b/hunts/confluence-space-permissions-opened.yaml
new file mode 100644
index 000000000..49fb6c060
--- /dev/null
+++ b/hunts/confluence-space-permissions-opened.yaml
@@ -0,0 +1,41 @@
+id: hunt-confluence-space-permissions-opened
+name: Documentation space opened to all authenticated users
+description: |
+ Space permissions change as teams grow. Opening a restricted space to every authenticated user exposes runbooks and architecture to anyone with a login.
+version: 1.0.0
+severity: medium
+category: application
+tags:
+ - mitre.attack.t1213.001
+ - hunt.saas
+log_sources:
+ - confluence_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a restricted documentation space opened to everyone?
+ indicators:
+ - field: "source"
+ equals: "confluence_audit"
+ - field: "action"
+ equals: "space_permissions_updated"
+ - field: "grantsAllAuthenticated"
+ equals: true
+ - field: "spaceSensitivity"
+ equals: "restricted"
+expected:
+ positive_incident_id: INC-HUNT-068
+ positive_template_id: hunt-confluence-space-permissions-opened
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-068-NEG
+ # The negative inverts grantsAllAuthenticated, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1213/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/container-exec-into-production-pod.yaml b/hunts/container-exec-into-production-pod.yaml
new file mode 100644
index 000000000..a808ec963
--- /dev/null
+++ b/hunts/container-exec-into-production-pod.yaml
@@ -0,0 +1,39 @@
+id: hunt-container-exec-into-production-pod
+name: Interactive shell opened inside a production pod
+description: |
+ Developers shell into development pods all day. The same action in production bypasses every deployment control the cluster has.
+version: 1.0.0
+severity: medium
+category: cloud
+tags:
+ - mitre.attack.t1609
+ - hunt.cloud
+log_sources:
+ - kubernetes_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did anyone open an interactive session inside a production pod?
+ indicators:
+ - field: "source"
+ equals: "kubernetes_audit"
+ - field: "subresource"
+ equals: "exec"
+ - field: "namespaceClass"
+ equals: "production"
+expected:
+ positive_incident_id: INC-HUNT-027
+ positive_template_id: hunt-container-exec-into-production-pod
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-027-NEG
+ # The negative inverts namespaceClass, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1609/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/dcsync-replication-rights.yaml b/hunts/dcsync-replication-rights.yaml
new file mode 100644
index 000000000..c9c855d0f
--- /dev/null
+++ b/hunts/dcsync-replication-rights.yaml
@@ -0,0 +1,41 @@
+id: hunt-dcsync-replication-rights
+name: Directory replication requested by a non-domain-controller account
+description: |
+ DCSync abuses the same replication rights domain controllers use legitimately. Replication between DCs is routine; the same request from an ordinary account is not.
+version: 1.0.0
+severity: critical
+category: identity
+tags:
+ - mitre.attack.t1003.006
+ - hunt.identity
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did an account that is not a domain controller request directory replication rights?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4662]
+ - field: "Properties"
+ regex: "1131f6a[ad]-9c07-11d1-f79f-00c04fc2dcd2"
+ - field: "SubjectIsDomainController"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-008
+ positive_template_id: hunt-dcsync-replication-rights
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-008-NEG
+ # The negative inverts SubjectIsDomainController, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1003/006/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/defender-exclusion-added.yaml b/hunts/defender-exclusion-added.yaml
new file mode 100644
index 000000000..2e8afd64f
--- /dev/null
+++ b/hunts/defender-exclusion-added.yaml
@@ -0,0 +1,42 @@
+id: hunt-defender-exclusion-added
+name: Anti-malware exclusion path added
+description: |
+ Exclusions are a normal part of endpoint policy. Set by the management agent they are policy; set by reg.exe on one host they are an attacker making room.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1562.001
+ - hunt.endpoint
+log_sources:
+ - sysmon
+ - defender
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an anti-malware exclusion added by something other than the management platform?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [13]
+ - field: "TargetObject"
+ regex: "(?i)\\\\Windows Defender\\\\Exclusions\\\\"
+ - field: "writtenByManagementAgent"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-020
+ positive_template_id: hunt-defender-exclusion-added
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-020-NEG
+ # The negative inverts writtenByManagementAgent, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1562/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/dns-over-https-to-unapproved-resolver.yaml b/hunts/dns-over-https-to-unapproved-resolver.yaml
new file mode 100644
index 000000000..afacf249e
--- /dev/null
+++ b/hunts/dns-over-https-to-unapproved-resolver.yaml
@@ -0,0 +1,40 @@
+id: hunt-dns-over-https-to-unapproved-resolver
+name: DNS-over-HTTPS to a resolver outside the approved set
+description: |
+ Encrypted DNS is normal and increasingly default. Using a resolver the organisation has not approved takes name resolution out of view of every DNS control.
+version: 1.0.0
+severity: medium
+category: network
+tags:
+ - mitre.attack.t1572
+ - hunt.network
+log_sources:
+ - proxy
+ - firewall
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any host resolve names through an unapproved encrypted resolver?
+ indicators:
+ - field: "source"
+ equals: "proxy"
+ - field: "url_path"
+ equals: "/dns-query"
+ - field: "resolverIsApproved"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-042
+ positive_template_id: hunt-dns-over-https-to-unapproved-resolver
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-042-NEG
+ # The negative inverts resolverIsApproved, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1572/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/docker-privileged-container-launched.yaml b/hunts/docker-privileged-container-launched.yaml
new file mode 100644
index 000000000..ce02da39d
--- /dev/null
+++ b/hunts/docker-privileged-container-launched.yaml
@@ -0,0 +1,44 @@
+id: hunt-docker-privileged-container-launched
+name: Privileged container started on a production node
+description: |
+ Production schedules pods continuously. A privileged one can reach the host's namespaces and is a container escape waiting for a reason.
+version: 1.0.0
+severity: high
+category: cloud
+tags:
+ - mitre.attack.t1610
+ - hunt.cloud
+log_sources:
+ - kubernetes_audit
+ - docker
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a privileged container scheduled onto a production node?
+ indicators:
+ - field: "source"
+ equals: "kubernetes_audit"
+ - field: "verb"
+ equals: "create"
+ - field: "resource"
+ equals: "pods"
+ - field: "podIsPrivileged"
+ equals: true
+ - field: "namespaceClass"
+ equals: "production"
+expected:
+ positive_incident_id: INC-HUNT-064
+ positive_template_id: hunt-docker-privileged-container-launched
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-064-NEG
+ # The negative inverts podIsPrivileged, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1610/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/duo-bypass-code-generated.yaml b/hunts/duo-bypass-code-generated.yaml
new file mode 100644
index 000000000..ee8ab85eb
--- /dev/null
+++ b/hunts/duo-bypass-code-generated.yaml
@@ -0,0 +1,40 @@
+id: hunt-duo-bypass-code-generated
+name: Multi-factor bypass code generated for an account
+description: |
+ Bypass codes exist because people lose phones, and the ticket is the record of that. A bypass with no ticket behind it is second-factor removal without a reason.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1556.006
+ - hunt.identity
+log_sources:
+ - duo
+ - jumpcloud
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a bypass code generated for an account that was not in a recorded lockout?
+ indicators:
+ - field: "source"
+ equals: "duo"
+ - field: "action"
+ equals: "bypass_create"
+ - field: "hasOpenLockoutTicket"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-059
+ positive_template_id: hunt-duo-bypass-code-generated
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-059-NEG
+ # The negative inverts hasOpenLockoutTicket, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1556/006/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/email-forwarding-rule-external.yaml b/hunts/email-forwarding-rule-external.yaml
new file mode 100644
index 000000000..8be0f3da9
--- /dev/null
+++ b/hunts/email-forwarding-rule-external.yaml
@@ -0,0 +1,39 @@
+id: hunt-email-forwarding-rule-external
+name: Inbox rule forwarding mail to an external address
+description: |
+ Delegating mail internally is routine. Forwarding it to an address the organisation does not control is the standard follow-through after a mailbox compromise.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1114.003
+ - hunt.identity
+log_sources:
+ - m365_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an inbox rule created that forwards mail outside the organisation?
+ indicators:
+ - field: "source"
+ equals: "m365_audit"
+ - field: "Operation"
+ in: ["New-InboxRule", "Set-InboxRule"]
+ - field: "forwardsExternally"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-040
+ positive_template_id: hunt-email-forwarding-rule-external
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-040-NEG
+ # The negative inverts forwardsExternally, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1114/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/encoded-powershell-command.yaml b/hunts/encoded-powershell-command.yaml
new file mode 100644
index 000000000..b04ccff92
--- /dev/null
+++ b/hunts/encoded-powershell-command.yaml
@@ -0,0 +1,41 @@
+id: hunt-encoded-powershell-command
+name: PowerShell invoked with a base64-encoded command
+description: |
+ Management platforms run encoded PowerShell as a matter of course. The same command line under an ordinary user account rarely has a benign explanation.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1059.001
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was PowerShell run with an encoded command by a non-administrative account?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [1]
+ - field: "CommandLine"
+ regex: "(?i)-(e|en|enc|encodedcommand)\\s+[A-Za-z0-9+/=]{40,}"
+ - field: "userIsAdministrator"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-026
+ positive_template_id: hunt-encoded-powershell-command
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-026-NEG
+ # The negative inverts userIsAdministrator, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1059/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/gcp-audit-sink-removed.yaml b/hunts/gcp-audit-sink-removed.yaml
new file mode 100644
index 000000000..2ff6342e7
--- /dev/null
+++ b/hunts/gcp-audit-sink-removed.yaml
@@ -0,0 +1,39 @@
+id: hunt-gcp-audit-sink-removed
+name: Cloud logging sink deleted
+description: |
+ Sinks are managed as code in most estates. A sink deleted by a person is the cloud equivalent of clearing the event log.
+version: 1.0.0
+severity: critical
+category: cloud
+tags:
+ - mitre.attack.t1562.008
+ - hunt.cloud
+log_sources:
+ - gcp_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a logging sink deleted or redirected away from the security project?
+ indicators:
+ - field: "source"
+ equals: "gcp_audit"
+ - field: "methodName"
+ regex: "(?i)ConfigServiceV2\\.(Delete|Update)Sink$"
+ - field: "performedByPipeline"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-023
+ positive_template_id: hunt-gcp-audit-sink-removed
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-023-NEG
+ # The negative inverts performedByPipeline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1562/008/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/gcp-service-account-key-created.yaml b/hunts/gcp-service-account-key-created.yaml
new file mode 100644
index 000000000..0dc99cb88
--- /dev/null
+++ b/hunts/gcp-service-account-key-created.yaml
@@ -0,0 +1,39 @@
+id: hunt-gcp-service-account-key-created
+name: Long-lived service account key created
+description: |
+ A downloadable key is a credential that leaves the platform and never expires. Pipelines still mint them for legacy systems; a person minting one usually wants it locally.
+version: 1.0.0
+severity: high
+category: cloud
+tags:
+ - mitre.attack.t1098.001
+ - hunt.cloud
+log_sources:
+ - gcp_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a downloadable service account key created despite workload identity being available?
+ indicators:
+ - field: "source"
+ equals: "gcp_audit"
+ - field: "methodName"
+ regex: "(?i)\\.CreateServiceAccountKey$"
+ - field: "performedByPipeline"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-053
+ positive_template_id: hunt-gcp-service-account-key-created
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-053-NEG
+ # The negative inverts performedByPipeline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1098/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/github-actions-workflow-added-by-outside-collaborator.yaml b/hunts/github-actions-workflow-added-by-outside-collaborator.yaml
new file mode 100644
index 000000000..779ede093
--- /dev/null
+++ b/hunts/github-actions-workflow-added-by-outside-collaborator.yaml
@@ -0,0 +1,39 @@
+id: hunt-github-actions-workflow-added-by-outside-collaborator
+name: CI workflow file added or changed by an account outside the organisation
+description: |
+ A workflow file is code that runs with repository credentials. Members change them all the time; an outside account changing one is a supply-chain question.
+version: 1.0.0
+severity: high
+category: application
+tags:
+ - mitre.attack.t1195.001
+ - hunt.supplychain
+log_sources:
+ - github_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did an account outside the organisation modify a CI workflow?
+ indicators:
+ - field: "source"
+ equals: "github_audit"
+ - field: "action"
+ in: ["workflows.created", "workflows.updated"]
+ - field: "actorIsOrgMember"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-017
+ positive_template_id: hunt-github-actions-workflow-added-by-outside-collaborator
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-017-NEG
+ # The negative inverts actorIsOrgMember, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1195/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/gitlab-protected-branch-rule-removed.yaml b/hunts/gitlab-protected-branch-rule-removed.yaml
new file mode 100644
index 000000000..607664875
--- /dev/null
+++ b/hunts/gitlab-protected-branch-rule-removed.yaml
@@ -0,0 +1,39 @@
+id: hunt-gitlab-protected-branch-rule-removed
+name: Branch protection removed from a production repository
+description: |
+ Protection rules change as teams reorganise. Removing them from a repository that ships to production removes the review that stands between a commit and a customer.
+version: 1.0.0
+severity: high
+category: application
+tags:
+ - mitre.attack.t1195.001
+ - hunt.supplychain
+log_sources:
+ - gitlab_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was branch protection removed from a repository that deploys to production?
+ indicators:
+ - field: "source"
+ equals: "gitlab_audit"
+ - field: "action"
+ in: ["protected_branch_removed", "protected_branch_updated"]
+ - field: "repoDeploysToProduction"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-061
+ positive_template_id: hunt-gitlab-protected-branch-rule-removed
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-061-NEG
+ # The negative inverts repoDeploysToProduction, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1195/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/imperva-waf-blocked-then-allowed-same-path.yaml b/hunts/imperva-waf-blocked-then-allowed-same-path.yaml
new file mode 100644
index 000000000..31eaf3e8d
--- /dev/null
+++ b/hunts/imperva-waf-blocked-then-allowed-same-path.yaml
@@ -0,0 +1,39 @@
+id: hunt-imperva-waf-blocked-then-allowed-same-path
+name: Requests to one path blocked repeatedly then allowed
+description: |
+ A blocked attacker is the control working, and blocks alone are background noise on any public service. The transition to allowed on the same path is the moment a bypass was found.
+version: 1.0.0
+severity: high
+category: application
+tags:
+ - mitre.attack.t1190
+ - hunt.application
+log_sources:
+ - imperva
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did a source get blocked repeatedly on a path and then succeed on it?
+ indicators:
+ - field: "source"
+ equals: "imperva"
+ - field: "priorBlocksSamePath"
+ gte: 20
+ - field: "action"
+ equals: "allow"
+expected:
+ positive_incident_id: INC-HUNT-067
+ positive_template_id: hunt-imperva-waf-blocked-then-allowed-same-path
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-067-NEG
+ # The negative inverts action, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1190/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/impossible-travel-sign-in.yaml b/hunts/impossible-travel-sign-in.yaml
new file mode 100644
index 000000000..203b6213a
--- /dev/null
+++ b/hunts/impossible-travel-sign-in.yaml
@@ -0,0 +1,42 @@
+id: hunt-impossible-travel-sign-in
+name: Successful sign-ins from locations too far apart to travel between
+description: |
+ A VPN egress in another region produces the same geography as impossible travel and is the single largest source of false positives in this pattern.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1078.004
+ - hunt.identity
+log_sources:
+ - okta
+ - azure_ad
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one account sign in successfully from two locations that could not both be true?
+ indicators:
+ - field: "source"
+ equals: "okta"
+ - field: "outcome"
+ equals: "SUCCESS"
+ - field: "impliedTravelSpeedKph"
+ gte: 900
+ - field: "viaCorporateVpn"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-054
+ positive_template_id: hunt-impossible-travel-sign-in
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-054-NEG
+ # The negative inverts viaCorporateVpn, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1078/004/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/jamf-device-unenrolled.yaml b/hunts/jamf-device-unenrolled.yaml
new file mode 100644
index 000000000..43ab0099c
--- /dev/null
+++ b/hunts/jamf-device-unenrolled.yaml
@@ -0,0 +1,39 @@
+id: hunt-jamf-device-unenrolled
+name: Managed device removed from device management
+description: |
+ Devices are retired constantly. A user unenrolling their own machine removes every control and every source of telemetry from it at once.
+version: 1.0.0
+severity: medium
+category: endpoint
+tags:
+ - mitre.attack.t1562.001
+ - hunt.endpoint
+log_sources:
+ - jamf
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a device unenrolled from management by someone other than IT?
+ indicators:
+ - field: "source"
+ equals: "jamf"
+ - field: "event"
+ equals: "ComputerRemoved"
+ - field: "actorIsItAdmin"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-057
+ positive_template_id: hunt-jamf-device-unenrolled
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-057-NEG
+ # The negative inverts actorIsItAdmin, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1562/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/k8s-audit-policy-weakened.yaml b/hunts/k8s-audit-policy-weakened.yaml
new file mode 100644
index 000000000..e6341027d
--- /dev/null
+++ b/hunts/k8s-audit-policy-weakened.yaml
@@ -0,0 +1,41 @@
+id: hunt-k8s-audit-policy-weakened
+name: Kubernetes audit or admission policy removed
+description: |
+ GitOps controllers reconcile webhook configurations constantly. A human deleting one removes the control that would have refused the next pod.
+version: 1.0.0
+severity: high
+category: cloud
+tags:
+ - mitre.attack.t1562.001
+ - hunt.cloud
+log_sources:
+ - kubernetes_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an admission or audit policy deleted from the cluster?
+ indicators:
+ - field: "source"
+ equals: "kubernetes_audit"
+ - field: "verb"
+ in: ["delete", "patch"]
+ - field: "resource"
+ in: ["validatingwebhookconfigurations", "auditsinks"]
+ - field: "performedByPipeline"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-024
+ positive_template_id: hunt-k8s-audit-policy-weakened
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-024-NEG
+ # The negative inverts performedByPipeline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1562/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/k8s-cluster-admin-binding.yaml b/hunts/k8s-cluster-admin-binding.yaml
new file mode 100644
index 000000000..80432e365
--- /dev/null
+++ b/hunts/k8s-cluster-admin-binding.yaml
@@ -0,0 +1,41 @@
+id: hunt-k8s-cluster-admin-binding
+name: Cluster-admin role bound to a service account
+description: |
+ Operators hold cluster-admin as people, with named accounts and an audit trail. A service account holding it means any workload that can read its token owns the cluster.
+version: 1.0.0
+severity: critical
+category: cloud
+tags:
+ - mitre.attack.t1078.003
+ - hunt.cloud
+log_sources:
+ - kubernetes_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was cluster-admin bound to a service account rather than to a person?
+ indicators:
+ - field: "source"
+ equals: "kubernetes_audit"
+ - field: "verb"
+ in: ["create", "update"]
+ - field: "roleRefName"
+ equals: "cluster-admin"
+ - field: "subjectKind"
+ equals: "ServiceAccount"
+expected:
+ positive_incident_id: INC-HUNT-051
+ positive_template_id: hunt-k8s-cluster-admin-binding
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-051-NEG
+ # The negative inverts subjectKind, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1078/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/kerberoast-rc4-service-ticket.yaml b/hunts/kerberoast-rc4-service-ticket.yaml
new file mode 100644
index 000000000..1a6f738a6
--- /dev/null
+++ b/hunts/kerberoast-rc4-service-ticket.yaml
@@ -0,0 +1,41 @@
+id: hunt-kerberoast-rc4-service-ticket
+name: Kerberos service ticket requested with RC4 encryption
+description: |
+ Service tickets issued with RC4 (0x17) rather than AES are the precondition for offline Kerberoasting. Modern domains issue AES by default, so a downgrade on a non-krbtgt service principal is worth a look.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1558.003
+ - hunt.identity
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any account request a Kerberos service ticket with downgraded RC4 encryption?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4769]
+ - field: "TicketEncryptionType"
+ equals: "0x17"
+ - field: "ServiceName"
+ regex: "^(?!krbtgt).+"
+expected:
+ positive_incident_id: INC-HUNT-006
+ positive_template_id: hunt-kerberoast-rc4-service-ticket
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-006-NEG
+ # The negative inverts TicketEncryptionType, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1558/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/linux-cron-added-outside-package-manager.yaml b/hunts/linux-cron-added-outside-package-manager.yaml
new file mode 100644
index 000000000..114c6dc86
--- /dev/null
+++ b/hunts/linux-cron-added-outside-package-manager.yaml
@@ -0,0 +1,41 @@
+id: hunt-linux-cron-added-outside-package-manager
+name: Cron entry created outside a package installation
+description: |
+ Package installs write cron entries routinely. A web-server account writing its own crontab is a different thing.
+version: 1.0.0
+severity: medium
+category: endpoint
+tags:
+ - mitre.attack.t1053.003
+ - hunt.endpoint
+log_sources:
+ - linux_auditd
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a cron file written by something other than the package manager?
+ indicators:
+ - field: "source"
+ equals: "linux_auditd"
+ - field: "syscall"
+ equals: "openat"
+ - field: "path"
+ regex: "^/(etc/cron|var/spool/cron)"
+ - field: "parentIsPackageManager"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-015
+ positive_template_id: hunt-linux-cron-added-outside-package-manager
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-015-NEG
+ # The negative inverts parentIsPackageManager, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1053/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/lsass-process-access.yaml b/hunts/lsass-process-access.yaml
new file mode 100644
index 000000000..d9ad49edb
--- /dev/null
+++ b/hunts/lsass-process-access.yaml
@@ -0,0 +1,41 @@
+id: hunt-lsass-process-access
+name: Non-standard process opened a handle to LSASS
+description: |
+ Credential dumping needs a handle to LSASS carrying VM_READ and QUERY_INFORMATION. The access mask is the discriminator: telemetry agents open LSASS constantly with far narrower rights.
+version: 1.0.0
+severity: critical
+category: endpoint
+tags:
+ - mitre.attack.t1003.001
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any process other than the known security agents open a handle to lsass.exe?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [10]
+ - field: "TargetImage"
+ iendswith: "\\lsass.exe"
+ - field: "GrantedAccess"
+ in: ["0x1010", "0x1410", "0x1438", "0x143a"]
+expected:
+ positive_incident_id: INC-HUNT-007
+ positive_template_id: hunt-lsass-process-access
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-007-NEG
+ # The negative inverts GrantedAccess, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1003/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/mass-file-rename-extension.yaml b/hunts/mass-file-rename-extension.yaml
new file mode 100644
index 000000000..2898c2665
--- /dev/null
+++ b/hunts/mass-file-rename-extension.yaml
@@ -0,0 +1,41 @@
+id: hunt-mass-file-rename-extension
+name: Large number of files renamed to a single new extension
+description: |
+ Batch processing renames thousands of files legitimately. An extension no installed application claims is what distinguishes encryption from a render farm.
+version: 1.0.0
+severity: critical
+category: endpoint
+tags:
+ - mitre.attack.t1486
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one process rename a large number of files to the same unfamiliar extension?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [11]
+ - field: "filesRenamedLast5Min"
+ gte: 500
+ - field: "extensionIsKnownApplication"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-046
+ positive_template_id: hunt-mass-file-rename-extension
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-046-NEG
+ # The negative inverts extensionIsKnownApplication, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1486/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/netskope-unsanctioned-app-upload.yaml b/hunts/netskope-unsanctioned-app-upload.yaml
new file mode 100644
index 000000000..823a16b51
--- /dev/null
+++ b/hunts/netskope-unsanctioned-app-upload.yaml
@@ -0,0 +1,42 @@
+id: hunt-netskope-unsanctioned-app-upload
+name: Upload to an application the organisation has not sanctioned
+description: |
+ Cloud uploads are how work happens. The sanction state of the destination is the organisation's own statement about where its data may go.
+version: 1.0.0
+severity: medium
+category: data-exfil
+tags:
+ - mitre.attack.t1567.002
+ - hunt.data
+log_sources:
+ - netskope
+ - cloudflare_zt
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did anyone upload data to an unsanctioned cloud application?
+ indicators:
+ - field: "source"
+ equals: "netskope"
+ - field: "activity"
+ equals: "Upload"
+ - field: "appSanctionState"
+ equals: "unsanctioned"
+ - field: "upload_bytes"
+ gte: 10000000
+expected:
+ positive_incident_id: INC-HUNT-058
+ positive_template_id: hunt-netskope-unsanctioned-app-upload
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-058-NEG
+ # The negative inverts appSanctionState, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1567/002/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/network-share-enumeration.yaml b/hunts/network-share-enumeration.yaml
new file mode 100644
index 000000000..7cc2f0ebe
--- /dev/null
+++ b/hunts/network-share-enumeration.yaml
@@ -0,0 +1,41 @@
+id: hunt-network-share-enumeration
+name: Host enumerated administrative shares across many peers
+description: |
+ Backup and management agents touch administrative shares on a handful of known hosts. Touching sixty in ten minutes is a sweep.
+version: 1.0.0
+severity: medium
+category: network
+tags:
+ - mitre.attack.t1135
+ - hunt.network
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one host connect to administrative shares on an unusual number of peers?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [5140]
+ - field: "ShareName"
+ regex: "(?i)\\\\\\\\\\*\\\\(ADMIN|C|IPC)\\$"
+ - field: "distinctPeersLast10Min"
+ gte: 20
+expected:
+ positive_incident_id: INC-HUNT-031
+ positive_template_id: hunt-network-share-enumeration
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-031-NEG
+ # The negative inverts distinctPeersLast10Min, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1135/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/newly-registered-domain-contact.yaml b/hunts/newly-registered-domain-contact.yaml
new file mode 100644
index 000000000..e18af1048
--- /dev/null
+++ b/hunts/newly-registered-domain-contact.yaml
@@ -0,0 +1,40 @@
+id: hunt-newly-registered-domain-contact
+name: Connection to a domain registered within the last week
+description: |
+ New domains appear constantly and most are ordinary. A new domain that no categorisation service recognises is the combination phishing and staging infrastructure share.
+version: 1.0.0
+severity: medium
+category: network
+tags:
+ - mitre.attack.t1071
+ - hunt.network
+log_sources:
+ - proxy
+ - dns
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any host contact a domain registered in the last seven days?
+ indicators:
+ - field: "source"
+ equals: "proxy"
+ - field: "domain_age_days"
+ lte: 7
+ - field: "destination_category"
+ equals: "uncategorised"
+expected:
+ positive_incident_id: INC-HUNT-043
+ positive_template_id: hunt-newly-registered-domain-contact
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-043-NEG
+ # The negative inverts destination_category, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1071/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/ntds-dit-shadow-copy.yaml b/hunts/ntds-dit-shadow-copy.yaml
new file mode 100644
index 000000000..919df8353
--- /dev/null
+++ b/hunts/ntds-dit-shadow-copy.yaml
@@ -0,0 +1,42 @@
+id: hunt-ntds-dit-shadow-copy
+name: Volume shadow copy created then NTDS.dit read
+description: |
+ Shadow copies are ordinary backup activity almost everywhere. On a domain controller they are the standard route to an offline copy of the AD database.
+version: 1.0.0
+severity: critical
+category: endpoint
+tags:
+ - mitre.attack.t1003.003
+ - hunt.endpoint
+log_sources:
+ - windows_security
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a shadow copy created on a domain controller and NTDS.dit read from it?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [1]
+ - field: "CommandLine"
+ regex: "(?i)(vssadmin|wmic).{0,60}(shadow|shadowcopy)"
+ - field: "IsDomainController"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-009
+ positive_template_id: hunt-ntds-dit-shadow-copy
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-009-NEG
+ # The negative inverts IsDomainController, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1003/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/oauth-refresh-token-long-lived-grant.yaml b/hunts/oauth-refresh-token-long-lived-grant.yaml
new file mode 100644
index 000000000..f89251588
--- /dev/null
+++ b/hunts/oauth-refresh-token-long-lived-grant.yaml
@@ -0,0 +1,42 @@
+id: hunt-oauth-refresh-token-long-lived-grant
+name: Long-lived OAuth grant issued to an application with no prior use
+description: |
+ offline_access is what turns a session into durable access. Consent without it expires; consent with it survives a password reset.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1550.001
+ - hunt.identity
+log_sources:
+ - azure_ad
+ - m365_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a long-lived token granted to an application nobody in the tenant had used before?
+ indicators:
+ - field: "source"
+ equals: "azure_ad"
+ - field: "operation"
+ equals: "Consent to application"
+ - field: "appFirstSeenInTenant"
+ equals: true
+ - field: "grantIncludesOfflineAccess"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-018
+ positive_template_id: hunt-oauth-refresh-token-long-lived-grant
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-018-NEG
+ # The negative inverts grantIncludesOfflineAccess, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1550/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/office-child-process-script-host.yaml b/hunts/office-child-process-script-host.yaml
new file mode 100644
index 000000000..49a66f713
--- /dev/null
+++ b/hunts/office-child-process-script-host.yaml
@@ -0,0 +1,41 @@
+id: hunt-office-child-process-script-host
+name: Office application spawned a script interpreter
+description: |
+ A document that starts a shell has almost always been made to. The parent alone is unremarkable: Office spawns printing and updater helpers constantly.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1566.001
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did an Office application start a scripting host or shell?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [1]
+ - field: "ParentImage"
+ regex: "(?i)\\\\(winword|excel|powerpnt|outlook)\\.exe$"
+ - field: "Image"
+ regex: "(?i)\\\\(powershell|cmd|wscript|cscript|mshta)\\.exe$"
+expected:
+ positive_incident_id: INC-HUNT-025
+ positive_template_id: hunt-office-child-process-script-host
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-025-NEG
+ # The negative inverts Image, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1566/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/okta-mfa-factor-reset-by-other-admin.yaml b/hunts/okta-mfa-factor-reset-by-other-admin.yaml
new file mode 100644
index 000000000..b4dc9dcd8
--- /dev/null
+++ b/hunts/okta-mfa-factor-reset-by-other-admin.yaml
@@ -0,0 +1,39 @@
+id: hunt-okta-mfa-factor-reset-by-other-admin
+name: MFA factor reset for an account by an administrator who has never done so before
+description: |
+ Help-desk staff reset MFA factors all day. An administrator doing it for the first time, especially against a high-value account, is the shape of a compromised or socially-engineered admin session.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1556.006
+ - hunt.identity
+log_sources:
+ - okta
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did an administrator reset another user's MFA factor for the first time?
+ indicators:
+ - field: "source"
+ equals: "okta"
+ - field: "eventType"
+ in: ["user.mfa.factor.reset_all", "user.mfa.factor.deactivate"]
+ - field: "actorFirstTimeResettingMfa"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-011
+ positive_template_id: hunt-okta-mfa-factor-reset-by-other-admin
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-011-NEG
+ # The negative inverts actorFirstTimeResettingMfa, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1556/006/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/osquery-agent-stopped-reporting.yaml b/hunts/osquery-agent-stopped-reporting.yaml
new file mode 100644
index 000000000..8fed951e7
--- /dev/null
+++ b/hunts/osquery-agent-stopped-reporting.yaml
@@ -0,0 +1,39 @@
+id: hunt-osquery-agent-stopped-reporting
+name: Endpoint agent stopped reporting while the host stayed reachable
+description: |
+ Laptops go offline all the time, which is why silence alone means nothing. Silence from a host that is still answering means something stopped the agent.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1562.001
+ - hunt.endpoint
+log_sources:
+ - osquery
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did an agent go silent on a host that is still answering on the network?
+ indicators:
+ - field: "source"
+ equals: "osquery"
+ - field: "agentSilentMinutes"
+ gte: 60
+ - field: "hostStillReachable"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-062
+ positive_template_id: hunt-osquery-agent-stopped-reporting
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-062-NEG
+ # The negative inverts hostStillReachable, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1562/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/password-spray-low-and-slow.yaml b/hunts/password-spray-low-and-slow.yaml
new file mode 100644
index 000000000..01dd9b6f7
--- /dev/null
+++ b/hunts/password-spray-low-and-slow.yaml
@@ -0,0 +1,42 @@
+id: hunt-password-spray-low-and-slow
+name: Many accounts failing authentication from one source with few attempts each
+description: |
+ Password spraying stays under per-account lockout thresholds by spreading across accounts, so the count of distinct accounts rather than the count of attempts is what surfaces it.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1110.003
+ - hunt.identity
+log_sources:
+ - okta
+ - azure_ad
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one source fail authentication against many distinct accounts, a couple of attempts each?
+ indicators:
+ - field: "source"
+ equals: "okta"
+ - field: "outcome"
+ equals: "FAILURE"
+ - field: "distinctAccountsFromIpLastHour"
+ gte: 30
+ - field: "attemptsPerAccount"
+ lte: 3
+expected:
+ positive_incident_id: INC-HUNT-055
+ positive_template_id: hunt-password-spray-low-and-slow
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-055-NEG
+ # The negative inverts distinctAccountsFromIpLastHour, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1110/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/privileged-group-membership-added.yaml b/hunts/privileged-group-membership-added.yaml
new file mode 100644
index 000000000..1f60d6a9f
--- /dev/null
+++ b/hunts/privileged-group-membership-added.yaml
@@ -0,0 +1,41 @@
+id: hunt-privileged-group-membership-added
+name: Account added to a privileged directory group
+description: |
+ Privileged groups do gain members legitimately through joiner-mover-leaver processes. Membership added outside that path is the thing worth catching.
+version: 1.0.0
+severity: high
+category: identity
+tags:
+ - mitre.attack.t1098
+ - hunt.identity
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an account added to a privileged group outside the approved provisioning path?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4728, 4732, 4756]
+ - field: "groupIsPrivileged"
+ equals: true
+ - field: "viaApprovedProvisioning"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-049
+ positive_template_id: hunt-privileged-group-membership-added
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-049-NEG
+ # The negative inverts viaApprovedProvisioning, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1098/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/rdp-from-workstation-to-workstation.yaml b/hunts/rdp-from-workstation-to-workstation.yaml
new file mode 100644
index 000000000..32f5699e0
--- /dev/null
+++ b/hunts/rdp-from-workstation-to-workstation.yaml
@@ -0,0 +1,43 @@
+id: hunt-rdp-from-workstation-to-workstation
+name: Remote desktop session between two workstations
+description: |
+ Support staff reach workstations from jump hosts. Workstation-to-workstation RDP bypasses that path and is one of the clearest lateral-movement shapes in Windows telemetry.
+version: 1.0.0
+severity: medium
+category: network
+tags:
+ - mitre.attack.t1021.001
+ - hunt.network
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did one workstation open a remote desktop session to another workstation?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4624]
+ - field: "LogonType"
+ in: [10]
+ - field: "sourceHostClass"
+ equals: "workstation"
+ - field: "targetHostClass"
+ equals: "workstation"
+expected:
+ positive_incident_id: INC-HUNT-033
+ positive_template_id: hunt-rdp-from-workstation-to-workstation
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-033-NEG
+ # The negative inverts sourceHostClass, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1021/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/run-key-persistence.yaml b/hunts/run-key-persistence.yaml
new file mode 100644
index 000000000..9b8e7f3a7
--- /dev/null
+++ b/hunts/run-key-persistence.yaml
@@ -0,0 +1,41 @@
+id: hunt-run-key-persistence
+name: Registry Run key written by a process that is not an installer
+description: |
+ Run keys are the most common persistence mechanism on Windows and also how a great deal of legitimate software starts. The writing process, not the key, is the signal.
+version: 1.0.0
+severity: medium
+category: endpoint
+tags:
+ - mitre.attack.t1547.001
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did a non-installer process write an autorun registry value?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [13]
+ - field: "TargetObject"
+ regex: "(?i)\\\\CurrentVersion\\\\Run"
+ - field: "ImageIsSignedInstaller"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-014
+ positive_template_id: hunt-run-key-persistence
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-014-NEG
+ # The negative inverts ImageIsSignedInstaller, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1547/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/s3-bucket-made-public.yaml b/hunts/s3-bucket-made-public.yaml
new file mode 100644
index 000000000..5f2bea79c
--- /dev/null
+++ b/hunts/s3-bucket-made-public.yaml
@@ -0,0 +1,39 @@
+id: hunt-s3-bucket-made-public
+name: Object storage bucket made publicly readable
+description: |
+ Bucket policies change constantly in an active estate. Whether the resulting policy grants anonymous read is the only part that matters.
+version: 1.0.0
+severity: high
+category: cloud
+tags:
+ - mitre.attack.t1530
+ - hunt.cloud
+log_sources:
+ - aws_cloudtrail
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a bucket policy changed to allow public access?
+ indicators:
+ - field: "source"
+ equals: "aws_cloudtrail"
+ - field: "eventName"
+ in: ["PutBucketPolicy", "PutBucketAcl", "DeletePublicAccessBlock"]
+ - field: "resultsInPublicAccess"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-047
+ positive_template_id: hunt-s3-bucket-made-public
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-047-NEG
+ # The negative inverts resultsInPublicAccess, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1530/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/saas-bulk-download.yaml b/hunts/saas-bulk-download.yaml
new file mode 100644
index 000000000..981288b14
--- /dev/null
+++ b/hunts/saas-bulk-download.yaml
@@ -0,0 +1,40 @@
+id: hunt-saas-bulk-download
+name: Bulk download from a SaaS application
+description: |
+ Everyone downloads files. Volume against the account's own baseline is what separates a working day from a copy of the document library.
+version: 1.0.0
+severity: high
+category: data-exfil
+tags:
+ - mitre.attack.t1567.002
+ - hunt.data
+log_sources:
+ - m365_audit
+ - gsuite_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any account download far more files in an hour than it normally does?
+ indicators:
+ - field: "source"
+ equals: "m365_audit"
+ - field: "Operation"
+ in: ["FileDownloaded", "FileSyncDownloadedFull"]
+ - field: "filesDownloadedLastHour"
+ gte: 200
+expected:
+ positive_incident_id: INC-HUNT-036
+ positive_template_id: hunt-saas-bulk-download
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-036-NEG
+ # The negative inverts filesDownloadedLastHour, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1567/002/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/salesforce-report-export-by-departing-user.yaml b/hunts/salesforce-report-export-by-departing-user.yaml
new file mode 100644
index 000000000..3d628e807
--- /dev/null
+++ b/hunts/salesforce-report-export-by-departing-user.yaml
@@ -0,0 +1,41 @@
+id: hunt-salesforce-report-export-by-departing-user
+name: Large CRM export by an account in its notice period
+description: |
+ Large exports are part of sales operations. The same export by someone whose last day is scheduled is the single most common insider-risk pattern.
+version: 1.0.0
+severity: high
+category: data-exfil
+tags:
+ - mitre.attack.t1567
+ - hunt.insider
+log_sources:
+ - salesforce
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did an account flagged as leaving export a large report?
+ indicators:
+ - field: "source"
+ equals: "salesforce"
+ - field: "EventType"
+ in: ["ReportExport"]
+ - field: "rowsExported"
+ gte: 5000
+ - field: "userIsInNoticePeriod"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-038
+ positive_template_id: hunt-salesforce-report-export-by-departing-user
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-038-NEG
+ # The negative inverts userIsInNoticePeriod, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1567/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/scheduled-task-from-temp-path.yaml b/hunts/scheduled-task-from-temp-path.yaml
new file mode 100644
index 000000000..857e684ad
--- /dev/null
+++ b/hunts/scheduled-task-from-temp-path.yaml
@@ -0,0 +1,40 @@
+id: hunt-scheduled-task-from-temp-path
+name: Scheduled task registered to run a binary from a temporary directory
+description: |
+ Scheduled tasks are how most software updates itself. A task whose action lives in a world-writable temporary directory is not how packaged software installs.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1053.005
+ - hunt.endpoint
+log_sources:
+ - windows_security
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a scheduled task created whose action points at a user-writable temporary path?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [4698]
+ - field: "TaskCommand"
+ regex: "(?i)\\\\(temp|tmp|appdata\\\\local\\\\temp|users\\\\public)\\\\"
+expected:
+ positive_incident_id: INC-HUNT-013
+ positive_template_id: hunt-scheduled-task-from-temp-path
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-013-NEG
+ # The negative inverts TaskCommand, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1053/005/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/shadow-copy-deletion.yaml b/hunts/shadow-copy-deletion.yaml
new file mode 100644
index 000000000..e708b0feb
--- /dev/null
+++ b/hunts/shadow-copy-deletion.yaml
@@ -0,0 +1,39 @@
+id: hunt-shadow-copy-deletion
+name: Volume shadow copies deleted
+description: |
+ Deleting every shadow copy is the step ransomware takes before encrypting, and it has almost no legitimate interactive use.
+version: 1.0.0
+severity: critical
+category: endpoint
+tags:
+ - mitre.attack.t1490
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Were shadow copies deleted, removing the local recovery path?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [1]
+ - field: "CommandLine"
+ regex: "(?i)(vssadmin|wmic).{0,40}(delete shadows|shadowcopy delete)"
+expected:
+ positive_incident_id: INC-HUNT-045
+ positive_template_id: hunt-shadow-copy-deletion
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-045-NEG
+ # The negative inverts CommandLine, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1490/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/slack-external-shared-channel-with-finance.yaml b/hunts/slack-external-shared-channel-with-finance.yaml
new file mode 100644
index 000000000..6c32b151e
--- /dev/null
+++ b/hunts/slack-external-shared-channel-with-finance.yaml
@@ -0,0 +1,39 @@
+id: hunt-slack-external-shared-channel-with-finance
+name: External shared channel opened against a sensitive internal channel
+description: |
+ Shared channels with suppliers are ordinary. Connecting one to a restricted channel exposes whatever has already been said in it.
+version: 1.0.0
+severity: medium
+category: application
+tags:
+ - mitre.attack.t1199
+ - hunt.saas
+log_sources:
+ - slack_audit
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a channel holding sensitive discussion shared with an external organisation?
+ indicators:
+ - field: "source"
+ equals: "slack_audit"
+ - field: "action"
+ equals: "external_shared_channel_connected"
+ - field: "channelSensitivity"
+ equals: "restricted"
+expected:
+ positive_incident_id: INC-HUNT-060
+ positive_template_id: hunt-slack-external-shared-channel-with-finance
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-060-NEG
+ # The negative inverts channelSensitivity, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1199/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/smb-admin-share-write-executable.yaml b/hunts/smb-admin-share-write-executable.yaml
new file mode 100644
index 000000000..d9b3ce20a
--- /dev/null
+++ b/hunts/smb-admin-share-write-executable.yaml
@@ -0,0 +1,41 @@
+id: hunt-smb-admin-share-write-executable
+name: Executable written to a remote administrative share
+description: |
+ Reading a binary from a share is ordinary software distribution. Writing one to another machine's ADMIN$ is how remote execution starts.
+version: 1.0.0
+severity: high
+category: network
+tags:
+ - mitre.attack.t1021.002
+ - hunt.network
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an executable written to another host's administrative share?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [5145]
+ - field: "RelativeTargetName"
+ regex: "(?i)\\.(exe|dll|ps1|bat|scr)$"
+ - field: "AccessMask"
+ regex: "(?i)WriteData|0x2"
+expected:
+ positive_incident_id: INC-HUNT-032
+ positive_template_id: hunt-smb-admin-share-write-executable
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-032-NEG
+ # The negative inverts AccessMask, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1021/002/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/snowflake-large-result-export.yaml b/hunts/snowflake-large-result-export.yaml
new file mode 100644
index 000000000..c9d1c1392
--- /dev/null
+++ b/hunts/snowflake-large-result-export.yaml
@@ -0,0 +1,41 @@
+id: hunt-snowflake-large-result-export
+name: Unusually large query result exported from the warehouse
+description: |
+ Nightly pipelines unload tens of millions of rows by design. The same volume under an interactive analyst account is a different event entirely.
+version: 1.0.0
+severity: high
+category: data-exfil
+tags:
+ - mitre.attack.t1567
+ - hunt.data
+log_sources:
+ - snowflake
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any user export an unusually large result set?
+ indicators:
+ - field: "source"
+ equals: "snowflake"
+ - field: "query_type"
+ in: ["UNLOAD", "COPY"]
+ - field: "rows_unloaded"
+ gte: 1000000
+ - field: "performedByPipeline"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-037
+ positive_template_id: hunt-snowflake-large-result-export
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-037-NEG
+ # The negative inverts performedByPipeline, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1567/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/ssh-key-added-to-authorized-keys.yaml b/hunts/ssh-key-added-to-authorized-keys.yaml
new file mode 100644
index 000000000..3c39e9128
--- /dev/null
+++ b/hunts/ssh-key-added-to-authorized-keys.yaml
@@ -0,0 +1,39 @@
+id: hunt-ssh-key-added-to-authorized-keys
+name: Public key appended to an account's authorized_keys
+description: |
+ Configuration management owns authorized_keys in most estates. A shell appending to it is the cheapest persistence on Linux.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1098.004
+ - hunt.endpoint
+log_sources:
+ - linux_auditd
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a key appended to authorized_keys by something other than configuration management?
+ indicators:
+ - field: "source"
+ equals: "linux_auditd"
+ - field: "path"
+ regex: "authorized_keys$"
+ - field: "parentIsConfigManagement"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-034
+ positive_template_id: hunt-ssh-key-added-to-authorized-keys
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-034-NEG
+ # The negative inverts parentIsConfigManagement, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1098/004/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/sudoers-modified-outside-config-management.yaml b/hunts/sudoers-modified-outside-config-management.yaml
new file mode 100644
index 000000000..55a62221a
--- /dev/null
+++ b/hunts/sudoers-modified-outside-config-management.yaml
@@ -0,0 +1,39 @@
+id: hunt-sudoers-modified-outside-config-management
+name: sudoers file changed outside configuration management
+description: |
+ sudoers is managed as code almost everywhere it matters. Edited by hand on one host, it is a privilege grant nobody reviewed.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1548.003
+ - hunt.endpoint
+log_sources:
+ - linux_auditd
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was the sudoers configuration modified by a shell rather than by configuration management?
+ indicators:
+ - field: "source"
+ equals: "linux_auditd"
+ - field: "path"
+ regex: "^/etc/sudoers"
+ - field: "parentIsConfigManagement"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-052
+ positive_template_id: hunt-sudoers-modified-outside-config-management
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-052-NEG
+ # The negative inverts parentIsConfigManagement, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1548/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/systemd-service-written-by-web-account.yaml b/hunts/systemd-service-written-by-web-account.yaml
new file mode 100644
index 000000000..ad48f4a0a
--- /dev/null
+++ b/hunts/systemd-service-written-by-web-account.yaml
@@ -0,0 +1,39 @@
+id: hunt-systemd-service-written-by-web-account
+name: systemd unit file created by a service account that serves traffic
+description: |
+ Configuration management writes unit files constantly. The account class is what separates a deployment from a web process that has been made to write one.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1543.002
+ - hunt.endpoint
+log_sources:
+ - linux_auditd
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did a web or application service account create a systemd unit?
+ indicators:
+ - field: "source"
+ equals: "linux_auditd"
+ - field: "path"
+ regex: "^/(etc|lib|usr/lib)/systemd/system/"
+ - field: "accountClass"
+ equals: "web"
+expected:
+ positive_incident_id: INC-HUNT-016
+ positive_template_id: hunt-systemd-service-written-by-web-account
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-016-NEG
+ # The negative inverts accountClass, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1543/002/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/tailscale-device-key-expiry-disabled.yaml b/hunts/tailscale-device-key-expiry-disabled.yaml
new file mode 100644
index 000000000..8de1e1381
--- /dev/null
+++ b/hunts/tailscale-device-key-expiry-disabled.yaml
@@ -0,0 +1,39 @@
+id: hunt-tailscale-device-key-expiry-disabled
+name: Device key expiry disabled on a tailnet node
+description: |
+ Network operators disable expiry on gateways deliberately. An ordinary user doing it to their own device turns a rotating credential into a permanent one.
+version: 1.0.0
+severity: medium
+category: network
+tags:
+ - mitre.attack.t1556
+ - hunt.network
+log_sources:
+ - tailscale
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was key expiry turned off on a node, making its credential permanent?
+ indicators:
+ - field: "source"
+ equals: "tailscale"
+ - field: "action"
+ equals: "device.keyExpiry.disable"
+ - field: "actorIsTailnetAdmin"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-035
+ positive_template_id: hunt-tailscale-device-key-expiry-disabled
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-035-NEG
+ # The negative inverts actorIsTailnetAdmin, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1556/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/timestomp-file-creation-time.yaml b/hunts/timestomp-file-creation-time.yaml
new file mode 100644
index 000000000..9eb13079b
--- /dev/null
+++ b/hunts/timestomp-file-creation-time.yaml
@@ -0,0 +1,39 @@
+id: hunt-timestomp-file-creation-time
+name: File creation time changed to an older value
+description: |
+ Build tooling rewrites timestamps forward routinely. Moving a creation time backwards, especially by months, is done to make a file look like it was always there.
+version: 1.0.0
+severity: medium
+category: endpoint
+tags:
+ - mitre.attack.t1070.006
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any process backdate a file's creation timestamp?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [2]
+ - field: "creationTimeMovedBackwardsDays"
+ gte: 1
+expected:
+ positive_incident_id: INC-HUNT-021
+ positive_template_id: hunt-timestomp-file-creation-time
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-021-NEG
+ # The negative inverts creationTimeMovedBackwardsDays, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1070/006/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/tor-exit-node-connection.yaml b/hunts/tor-exit-node-connection.yaml
new file mode 100644
index 000000000..ffc9aa629
--- /dev/null
+++ b/hunts/tor-exit-node-connection.yaml
@@ -0,0 +1,39 @@
+id: hunt-tor-exit-node-connection
+name: Outbound connection to a known anonymising relay
+description: |
+ Blocked attempts are the control working. An allowed connection to a relay means egress policy has a hole an operator should know about.
+version: 1.0.0
+severity: high
+category: network
+tags:
+ - mitre.attack.t1090.003
+ - hunt.network
+log_sources:
+ - firewall
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any internal host connect to an anonymising relay?
+ indicators:
+ - field: "source"
+ equals: "firewall"
+ - field: "action"
+ equals: "allow"
+ - field: "destination_is_anonymising_relay"
+ equals: true
+expected:
+ positive_incident_id: INC-HUNT-044
+ positive_template_id: hunt-tor-exit-node-connection
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-044-NEG
+ # The negative inverts action, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1090/003/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/vault-secret-mass-read.yaml b/hunts/vault-secret-mass-read.yaml
new file mode 100644
index 000000000..ed59877af
--- /dev/null
+++ b/hunts/vault-secret-mass-read.yaml
@@ -0,0 +1,39 @@
+id: hunt-vault-secret-mass-read
+name: One identity read an unusual number of distinct secrets
+description: |
+ A pipeline reads the handful of secrets it needs. Enumerating the store is what an operator does when they have a token and do not yet know what it unlocks.
+version: 1.0.0
+severity: high
+category: data-exfil
+tags:
+ - mitre.attack.t1552.007
+ - hunt.data
+log_sources:
+ - vault
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any identity read far more distinct secret paths in an hour than it normally does?
+ indicators:
+ - field: "source"
+ equals: "vault"
+ - field: "operation"
+ equals: "read"
+ - field: "distinct_paths_last_hour"
+ gte: 50
+expected:
+ positive_incident_id: INC-HUNT-012
+ positive_template_id: hunt-vault-secret-mass-read
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-012-NEG
+ # The negative inverts distinct_paths_last_hour, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1552/007/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/vpn-login-from-hosting-provider.yaml b/hunts/vpn-login-from-hosting-provider.yaml
new file mode 100644
index 000000000..35ee5667f
--- /dev/null
+++ b/hunts/vpn-login-from-hosting-provider.yaml
@@ -0,0 +1,39 @@
+id: hunt-vpn-login-from-hosting-provider
+name: Remote access session from a hosting provider address range
+description: |
+ Staff connect from home and from offices. A session from a cloud hosting range usually means the credential is being used from infrastructure rather than by a person at a desk.
+version: 1.0.0
+severity: medium
+category: network
+tags:
+ - mitre.attack.t1133
+ - hunt.network
+log_sources:
+ - vpn
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Did any remote access session originate from a datacentre rather than a residential or corporate network?
+ indicators:
+ - field: "source"
+ equals: "vpn"
+ - field: "status"
+ equals: "connected"
+ - field: "sourceAsnClass"
+ equals: "hosting"
+expected:
+ positive_incident_id: INC-HUNT-056
+ positive_template_id: hunt-vpn-login-from-hosting-provider
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-056-NEG
+ # The negative inverts sourceAsnClass, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1133/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/windows-event-log-cleared.yaml b/hunts/windows-event-log-cleared.yaml
new file mode 100644
index 000000000..b99ac0370
--- /dev/null
+++ b/hunts/windows-event-log-cleared.yaml
@@ -0,0 +1,39 @@
+id: hunt-windows-event-log-cleared
+name: Security event log cleared
+description: |
+ Re-imaging and provisioning clear logs legitimately. The same event outside a maintenance window has no ordinary explanation.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1070.001
+ - hunt.endpoint
+log_sources:
+ - windows_security
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was an event log cleared outside a maintenance window?
+ indicators:
+ - field: "source"
+ equals: "windows_security"
+ - field: "EventID"
+ in: [1102, 104]
+ - field: "duringMaintenanceWindow"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-019
+ positive_template_id: hunt-windows-event-log-cleared
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-019-NEG
+ # The negative inverts duringMaintenanceWindow, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1070/001/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/wmi-remote-process-creation.yaml b/hunts/wmi-remote-process-creation.yaml
new file mode 100644
index 000000000..16da655be
--- /dev/null
+++ b/hunts/wmi-remote-process-creation.yaml
@@ -0,0 +1,41 @@
+id: hunt-wmi-remote-process-creation
+name: Process created remotely over WMI
+description: |
+ Management servers drive WMI across the estate constantly. WMI-spawned processes arriving on an ordinary workstation are how lateral movement looks.
+version: 1.0.0
+severity: high
+category: endpoint
+tags:
+ - mitre.attack.t1047
+ - hunt.endpoint
+log_sources:
+ - sysmon
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a process created by the WMI provider host on a workstation?
+ indicators:
+ - field: "source"
+ equals: "sysmon"
+ - field: "EventID"
+ in: [1]
+ - field: "ParentImage"
+ iendswith: "\\wmiprvse.exe"
+ - field: "hostClass"
+ equals: "workstation"
+expected:
+ positive_incident_id: INC-HUNT-028
+ positive_template_id: hunt-wmi-remote-process-creation
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-028-NEG
+ # The negative inverts hostClass, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1047/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/hunts/zoom-recording-downloaded-externally.yaml b/hunts/zoom-recording-downloaded-externally.yaml
new file mode 100644
index 000000000..e76f61da9
--- /dev/null
+++ b/hunts/zoom-recording-downloaded-externally.yaml
@@ -0,0 +1,39 @@
+id: hunt-zoom-recording-downloaded-externally
+name: Meeting recording downloaded by an account outside the organisation
+description: |
+ Recordings are shared internally as a matter of course. An external download of one is a disclosure whether or not it was intended.
+version: 1.0.0
+severity: medium
+category: data-exfil
+tags:
+ - mitre.attack.t1213
+ - hunt.saas
+log_sources:
+ - zoom
+schedule:
+ enabled: true
+ interval_minutes: 60
+ jitter_seconds: 60
+hypothesis:
+ question: |
+ Was a recording downloaded by an external account?
+ indicators:
+ - field: "source"
+ equals: "zoom"
+ - field: "event"
+ equals: "recording.downloaded"
+ - field: "downloaderIsInternal"
+ equals: false
+expected:
+ positive_incident_id: INC-HUNT-063
+ positive_template_id: hunt-zoom-recording-downloaded-externally
+ min_match_score: 0.8
+ negative_incident_id: INC-HUNT-063-NEG
+ # The negative inverts downloaderIsInternal, the clause this hunt is about,
+ # and holds every other field constant. A negative that differed in the
+ # log source would pass without testing anything.
+references:
+ - https://attack.mitre.org/techniques/T1213/
+author: AiSOC
+created: '2026-09-27'
+modified: '2026-09-27'
diff --git a/scripts/check_hunt_scenarios.py b/scripts/check_hunt_scenarios.py
new file mode 100644
index 000000000..7d6dc1143
--- /dev/null
+++ b/scripts/check_hunt_scenarios.py
@@ -0,0 +1,323 @@
+#!/usr/bin/env python3
+"""A hunt's negative scenario must invert the clause the hunt is about.
+
+Why this exists
+---------------
+
+``services/agents/tests/test_hunt_corpus.py`` grades every hunt against a
+positive and a negative synthetic scenario, and a hunt passes when it fires on
+the first and not on the second. That is necessary and it is not sufficient,
+because the second half is trivially satisfiable. A negative scenario drawn
+from a different log source entirely will never fire on any hunt, so a corpus
+of those would show a perfect false-positive rate while testing nothing.
+
+This repository already knows how that goes. Roughly 600 of its 825 detection
+fixtures are synthesised from the rule they test, which makes replay circular,
+and over a hundred negative fixtures had never been replayed at all under a
+test whose name said they were. The way out both times was to make the corpus
+prove a property the grading could not fake.
+
+So the rule here is the one the adversarial injection corpus already uses for
+its clean twins: **a negative differs from its positive in exactly one
+indicator field**, and that field is not the one selecting the log source. A
+negative shaped that way is a near miss, which is the only kind of negative
+that tests a hunt's discrimination rather than its aim.
+
+What it checks
+--------------
+
+``no-negative``
+ A hunt declares no negative scenario, or declares one with no events. It
+ can only ever be graded in one direction.
+
+``negative-wrong-source``
+ The negative fails the hunt's log-source indicator, so it is an unrelated
+ event and the hunt would not have fired on it whatever its logic was.
+
+``negative-fires``
+ The negative matches every indicator. It is a positive with another name.
+
+``negative-too-far``
+ The negative differs from the positive in more than one indicator field.
+ Each extra difference is another reason the hunt did not fire, and with
+ two the scenario no longer says which clause did the work.
+
+``negative-not-a-near-miss``
+ The negative matches less than half the hunt's indicators. Technically a
+ non-firing event, practically an unrelated one.
+
+``positive-does-not-fire``
+ Restated from the corpus test so this gate is not vacuous when run alone.
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+import types
+from dataclasses import dataclass
+from pathlib import Path
+
+from gate_toolkit import repo_root, self_test_if_requested
+
+self_test_if_requested(__file__)
+
+_HUNTS = Path("hunts")
+_TELEMETRY = Path("services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl")
+_ENGINE = Path("services/agents/app/hunt/engine.py")
+_LOADER = Path("services/agents/app/hunt/loader.py")
+
+REQUIRED = (_HUNTS, _TELEMETRY, _ENGINE, _LOADER)
+
+#: Indicator fields that select *which telemetry* rather than *what happened*.
+#: A negative that differs here is a different kind of event, not a near miss.
+_SOURCE_FIELDS = frozenset({"source", "log_source"})
+
+#: A negative matching fewer than this fraction of a hunt's indicators is not
+#: a near miss. Half is deliberately generous: the binding rule is the
+#: one-field difference below, and this catches a scenario that drifted.
+_NEAR_MISS_FLOOR = 0.5
+
+
+@dataclass
+class Finding:
+ kind: str
+ hunt_id: str
+ detail: str
+
+ def __str__(self) -> str:
+ return f" [{self.kind}] {self.hunt_id}: {self.detail}"
+
+
+def _load_engine(root: Path): # noqa: ANN202 - returns (loader, engine) modules
+ """Load the hunt engine and loader by path, leaving ``sys.modules`` clean.
+
+ The engine is the production matcher, so this gate asks the same question
+ the scheduler would. Re-implementing indicator matching here would make
+ the gate agree with itself rather than with the engine, which is the exact
+ circularity it exists to detect.
+
+ The relative import inside ``engine.py`` needs its package to resolve, so
+ two namespace modules are registered and then removed. Leaving a synthetic
+ name behind has previously broken 21 unrelated tests in this repository
+ while every test in its own file passed.
+ """
+ import importlib.util # noqa: PLC0415 - only needed on this path
+
+ created: list[str] = []
+ for name in ("app", "app.hunt"):
+ if name not in sys.modules:
+ module = types.ModuleType(name)
+ module.__path__ = [] # type: ignore[attr-defined]
+ sys.modules[name] = module
+ created.append(name)
+
+ def _load(rel: Path, alias: str): # noqa: ANN202
+ spec = importlib.util.spec_from_file_location(alias, root / rel)
+ if spec is None or spec.loader is None:
+ raise RuntimeError(f"cannot load {rel}")
+ module = importlib.util.module_from_spec(spec)
+ sys.modules[alias] = module
+ spec.loader.exec_module(module)
+ return module
+
+ try:
+ loader = _load(_LOADER, "app.hunt.loader")
+ engine = _load(_ENGINE, "app.hunt.engine")
+ return loader, engine
+ finally:
+ for name in ("app.hunt.engine", "app.hunt.loader", *created):
+ sys.modules.pop(name, None)
+
+
+def _events_by_incident(root: Path) -> dict[str, list[dict]]:
+ out: dict[str, list[dict]] = {}
+ with (root / _TELEMETRY).open(encoding="utf-8") as handle:
+ for line_no, raw in enumerate(handle, 1):
+ line = raw.strip()
+ if not line:
+ continue
+ try:
+ event = json.loads(line)
+ except json.JSONDecodeError as exc:
+ raise SystemExit(f"{_TELEMETRY}:{line_no} is not valid JSON: {exc}") from exc
+ out.setdefault(str(event.get("incident_id")), []).append(event)
+ return out
+
+
+def judge(root: Path) -> tuple[list[Finding], dict[str, int]]:
+ loader_mod, engine_mod = _load_engine(root)
+ corpus = loader_mod.HuntCorpus(root / _HUNTS)
+ corpus.reload()
+ hunts = corpus.list()
+ events = _events_by_incident(root)
+
+ findings: list[Finding] = []
+ stats = {"hunts": len(hunts), "one_field": 0, "graded_both_ways": 0}
+
+ if not hunts:
+ raise SystemExit(f"no hunts loaded from {root / _HUNTS}; refusing to report a corpus clean")
+
+ for hunt in hunts:
+ indicators = hunt.hypothesis.indicators
+ pos_id = hunt.expected.positive_incident_id
+ neg_id = hunt.expected.negative_incident_id
+ pos_events = events.get(pos_id or "", [])
+ neg_events = events.get(neg_id or "", [])
+
+ if not pos_events:
+ findings.append(Finding("positive-does-not-fire", hunt.id, f"no events for positive scenario {pos_id!r}"))
+ continue
+ if not neg_id or not neg_events:
+ findings.append(
+ Finding(
+ "no-negative",
+ hunt.id,
+ "declares no negative scenario, so it can only be graded in one direction",
+ )
+ )
+ continue
+
+ # The best positive event, so a multi-event scenario is judged on the
+ # one the hunt is about rather than on an incidental first row.
+ best_pos = max(pos_events, key=lambda e: sum(1 for i in indicators if engine_mod._indicator_matches(e, i)))
+ pos_hits = [engine_mod._indicator_matches(best_pos, i) for i in indicators]
+ if not all(pos_hits):
+ missed = [i.field for i, hit in zip(indicators, pos_hits, strict=True) if not hit]
+ findings.append(Finding("positive-does-not-fire", hunt.id, f"positive scenario fails indicator(s) on {missed}"))
+ continue
+
+ # The negative event closest to firing: if any negative event is a
+ # near miss the scenario is doing its job, and judging the weakest one
+ # would report a corpus worse than it is.
+ best_neg = max(neg_events, key=lambda e: sum(1 for i in indicators if engine_mod._indicator_matches(e, i)))
+ neg_hits = [engine_mod._indicator_matches(best_neg, i) for i in indicators]
+ score = sum(neg_hits) / len(indicators) if indicators else 0.0
+
+ if all(neg_hits):
+ findings.append(
+ Finding("negative-fires", hunt.id, "the negative scenario matches every indicator; it is a positive under another name")
+ )
+ continue
+
+ source_failures = [i.field for i, hit in zip(indicators, neg_hits, strict=True) if not hit and i.field in _SOURCE_FIELDS]
+ if source_failures:
+ findings.append(
+ Finding(
+ "negative-wrong-source",
+ hunt.id,
+ f"the negative fails the log-source indicator on {source_failures}, so it is an "
+ f"unrelated event rather than a near miss",
+ )
+ )
+ continue
+
+ if score < _NEAR_MISS_FLOOR:
+ findings.append(
+ Finding(
+ "negative-not-a-near-miss",
+ hunt.id,
+ f"the negative matches only {score:.0%} of the hunt's indicators; it is not testing the discriminating clause",
+ )
+ )
+ continue
+
+ # The binding rule. Compare on indicator fields only: incidental
+ # context (a different hostname, a different user) is fine and
+ # expected, but every *indicator* the hunt names must hold constant
+ # except the one the scenario is about.
+ differing = sorted({i.field for i in indicators if _get(best_pos, i.field) != _get(best_neg, i.field)})
+ if len(differing) > 1:
+ findings.append(
+ Finding(
+ "negative-too-far",
+ hunt.id,
+ f"the negative differs from the positive on {len(differing)} indicator fields ({differing}); "
+ f"with more than one difference the scenario no longer says which clause did the work",
+ )
+ )
+ continue
+
+ stats["one_field"] += 1
+ stats["graded_both_ways"] += 1
+
+ return findings, stats
+
+
+def _get(event: dict, dotted: str): # noqa: ANN202
+ cur = event
+ for part in dotted.split("."):
+ if isinstance(cur, dict) and part in cur:
+ cur = cur[part]
+ else:
+ return None
+ return cur
+
+
+def _self_test() -> int:
+ """Prove the gate detects each violation, against the committed corpus."""
+ from gate_toolkit import self_test_main # noqa: PLC0415
+
+ root = repo_root()
+ extra: list[tuple[str, bool]] = []
+
+ findings, stats = judge(root)
+ extra.append((f"the committed corpus of {stats['hunts']} hunts has no findings", not findings))
+ for finding in findings:
+ print(f" {finding}")
+
+ loader_mod, engine_mod = _load_engine(root)
+ corpus = loader_mod.HuntCorpus(root / _HUNTS)
+ corpus.reload()
+ sample = corpus.list()[0]
+ events = _events_by_incident(root)
+
+ # A negative drawn from another log source is the tautology this exists
+ # to catch. Built here rather than committed, so the corpus stays clean.
+ pos = events[sample.expected.positive_incident_id][0]
+ unrelated = {**pos, "source": "a-log-source-this-hunt-never-reads"}
+ indicators = sample.hypothesis.indicators
+ fails_source = any(not engine_mod._indicator_matches(unrelated, i) for i in indicators if i.field in _SOURCE_FIELDS)
+ extra.append(("a negative from an unrelated log source is recognisable as such", fails_source))
+
+ # And a negative identical to the positive fires, which is the other end.
+ fires = all(engine_mod._indicator_matches(pos, i) for i in indicators)
+ extra.append(("a negative identical to the positive would match every indicator", fires))
+
+ extra.append(("every graded hunt differs from its negative in exactly one indicator field", stats["one_field"] == stats["hunts"]))
+
+ return self_test_main(Path(__file__).name, ["--check"], extra=extra)
+
+
+def main(argv: list[str] | None = None) -> int:
+ args = list(argv if argv is not None else sys.argv[1:])
+ if "--self-test" in args:
+ return _self_test()
+
+ root = repo_root()
+ missing = [str(rel) for rel in REQUIRED if not (root / rel).exists()]
+ if missing:
+ print("check_hunt_scenarios: refusing to render a verdict — these paths are missing:")
+ for path in missing:
+ print(f" {path}")
+ return 2
+
+ findings, stats = judge(root)
+ if findings:
+ print(f"check_hunt_scenarios: {len(findings)} finding(s)\n")
+ for finding in findings:
+ print(finding)
+ print("\nA negative scenario that fails for an unrelated reason grades a hunt's aim rather than")
+ print("its discrimination, and a corpus of those reports a perfect false-positive rate while")
+ print("testing nothing.")
+ return 1
+
+ print(
+ f"check_hunt_scenarios: OK — {stats['hunts']} hunts, all graded in both directions, "
+ f"every negative inverting exactly one indicator field of its positive."
+ )
+ return 0
+
+
+if __name__ == "__main__":
+ raise SystemExit(main())
diff --git a/scripts/run_hunt_evals.py b/scripts/run_hunt_evals.py
new file mode 100644
index 000000000..29bcff391
--- /dev/null
+++ b/scripts/run_hunt_evals.py
@@ -0,0 +1,179 @@
+#!/usr/bin/env python3
+"""Grade the ``hunts/`` corpus. A thin wrapper over the unified runner.
+
+``hunts/README.md`` told contributors to run this script before opening a PR.
+It had never existed, so the one instruction in that document a contributor
+was most likely to follow failed at the shell. Gap-closure Phase 8.4.
+
+Deliberately a wrapper rather than an implementation. ``scripts/run_evals.py``
+already owns a ``hunt_corpus`` suite, with the floors CI gates on, and a
+second scorer here would drift from the one the scoreboard publishes. That has
+happened before in this repository: ``scripts/run_model_matrix.py`` is a thin
+wrapper for the same reason, and a test asserts it defines no scoring of its
+own. The same assertion covers this script.
+
+So this adds exactly two things over ``run_evals.py --suite hunt_corpus``: a
+name a contributor can guess, and output shaped for the question they are
+actually asking, which is "does the hunt I just wrote fire on its positive and
+stay quiet on its negative".
+
+Usage:
+ python scripts/run_hunt_evals.py # grade the whole corpus
+ python scripts/run_hunt_evals.py --hunt hunt-x # one hunt, with detail
+ python scripts/run_hunt_evals.py --json # machine-readable
+"""
+
+from __future__ import annotations
+
+import argparse
+import ast
+import json
+import sys
+from pathlib import Path
+
+from gate_toolkit import repo_root
+
+#: Resolved through git rather than from ``__file__``. Two levels above this
+#: script is whatever happens to be there, which is a different tree whenever
+#: the script is copied, vendored or run from a scratch checkout.
+_REPO_ROOT = repo_root()
+_RUN_EVALS = _REPO_ROOT / "scripts" / "run_evals.py"
+
+
+def _grade() -> dict:
+ """Run the corpus suite through the runner that owns the floors."""
+ sys.path.insert(0, str(_REPO_ROOT / "services" / "agents"))
+ try:
+ from tests.test_hunt_corpus import evaluate_hunt_corpus # noqa: PLC0415
+ except ImportError as exc:
+ print(
+ f"error: could not import the hunt corpus grader ({exc}).\n"
+ f"Run from the repository root with the agents service's test dependencies installed:\n"
+ f" pip install pyyaml pydantic\n",
+ file=sys.stderr,
+ )
+ raise SystemExit(2) from exc
+
+ result = evaluate_hunt_corpus()
+ return {
+ "hunts": result.hunts_total,
+ "positives_expected": result.positives_expected,
+ "positives_caught": result.positives_caught,
+ "positive_rate": round(result.positive_rate, 4),
+ "negatives_expected": result.negatives_expected,
+ "false_positives": result.false_positives,
+ "false_positive_rate": round(result.false_positive_rate, 4),
+ "misses": result.misses,
+ "false_positive_details": result.false_positive_details,
+ "orphan_incident_ids": result.orphan_incident_ids,
+ }
+
+
+def _report(data: dict, only: str | None) -> int:
+ misses = [m for m in data["misses"] if only is None or m.get("hunt_id") == only]
+ fps = [f for f in data["false_positive_details"] if only is None or f.get("hunt_id") == only]
+
+ print(f"hunt corpus: {data['hunts']} hunts")
+ print(f" positives {data['positives_caught']}/{data['positives_expected']} fired (rate {data['positive_rate']:.3f}, floor 1.000)")
+ print(
+ f" negatives {data['false_positives']}/{data['negatives_expected']} fired "
+ f"(rate {data['false_positive_rate']:.3f}, ceiling 0.000)"
+ )
+
+ if misses:
+ print("\nThese hunts did not fire on their positive scenario:")
+ for miss in misses:
+ reason = miss.get("reason")
+ if reason == "no_matching_events":
+ print(f" - {miss['hunt_id']}: scenario {miss['scenario']} has no events in the telemetry corpus")
+ else:
+ print(
+ f" - {miss['hunt_id']}: best score {miss.get('match_score')} < "
+ f"{miss.get('threshold')} over {miss.get('events_scanned')} event(s)"
+ )
+
+ if fps:
+ print("\nThese hunts fired on their negative scenario:")
+ for fp in fps:
+ print(f" - {fp['hunt_id']}: {fp['findings']} finding(s) on {fp['scenario']} (score {fp.get('match_score')})")
+
+ if data["orphan_incident_ids"]:
+ print("\nTelemetry left behind by a renamed or deleted hunt:")
+ for orphan in data["orphan_incident_ids"]:
+ print(f" - {orphan}")
+
+ ok = not misses and not fps and not data["orphan_incident_ids"]
+ if only is not None and not ok:
+ print(f"\n{only}: FAILED")
+ return 1
+ if ok:
+ print("\nOK. Every hunt fires on its positive scenario and none fire on their negative.")
+ print("Run scripts/check_hunt_scenarios.py as well: it checks that each negative inverts")
+ print("the clause its hunt is about, which this grading cannot tell on its own.")
+ return 0
+ print("\nFAILED.")
+ return 1
+
+
+def _self_test() -> int:
+ """Assert this script defines no scoring of its own.
+
+ The failure mode a wrapper exists to prevent is becoming a second
+ implementation, at which point two numbers describe the same corpus and
+ disagree. Read from this file's own source rather than asserted in prose.
+ """
+ source = Path(__file__).read_text(encoding="utf-8")
+ tree = ast.parse(source)
+ ok = True
+
+ imported: set[str] = set()
+ modules: set[str] = set()
+ for node in ast.walk(tree):
+ if isinstance(node, ast.ImportFrom):
+ modules.add(node.module or "")
+ imported.update(alias.name for alias in node.names)
+ elif isinstance(node, ast.Import):
+ modules.update(alias.name for alias in node.names)
+
+ delegates = "evaluate_hunt_corpus" in imported
+ print(f" {'PASS' if delegates else 'FAIL'} delegates to the corpus grader run_evals.py uses")
+ ok &= delegates
+
+ # A scorer would need the matcher and the corpus loader. Checked as
+ # *imports* rather than as substrings: the first version of this test
+ # searched the file text and failed on the names inside its own assertion
+ # message, which is a small instance of the thing it is checking for.
+ scoring_names = {"HuntEngine", "HuntCorpus", "HuntDefinition", "_indicator_matches"}
+ scoring_modules = {m for m in modules if "hunt.engine" in m or "hunt.loader" in m}
+ cannot_score = not (imported & scoring_names) and not scoring_modules
+ print(f" {'PASS' if cannot_score else 'FAIL'} imports neither the hunt engine nor the loader, so it cannot score")
+ ok &= cannot_score
+
+ runner_exists = _RUN_EVALS.exists()
+ print(f" {'PASS' if runner_exists else 'FAIL'} the runner it defers to exists at scripts/run_evals.py")
+ ok &= runner_exists
+
+ print()
+ print(f"run_hunt_evals.py: self-test {'OK' if ok else 'FAILED'}")
+ return 0 if ok else 1
+
+
+def main(argv: list[str] | None = None) -> int:
+ parser = argparse.ArgumentParser(description="Grade the hunts/ corpus against its synthetic scenarios.")
+ parser.add_argument("--hunt", help="Report on one hunt id only, for instance hunt-lolbin-rundll32-network.")
+ parser.add_argument("--json", action="store_true", help="Print the raw grading result as JSON.")
+ parser.add_argument("--self-test", action="store_true", help="Assert this wrapper defines no scoring of its own.")
+ args = parser.parse_args(argv)
+
+ if args.self_test:
+ return _self_test()
+
+ data = _grade()
+ if args.json:
+ print(json.dumps(data, indent=2))
+ return 0 if data["positive_rate"] >= 1.0 and data["false_positive_rate"] <= 0.0 else 1
+ return _report(data, args.hunt)
+
+
+if __name__ == "__main__":
+ raise SystemExit(main())
diff --git a/services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl b/services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl
index 0dfe150f8..e8b5a2983 100644
--- a/services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl
+++ b/services/agents/tests/eval_data/synthetic_hunt_telemetry.jsonl
@@ -3,13 +3,139 @@
{"incident_id": "INC-HUNT-001-NEG", "template_id": "hunt-svc-account-after-hours-neg", "event_index": 0, "source": "windows_security", "channel": "Security", "Provider": "Microsoft-Windows-Security-Auditing", "EventID": 4624, "Computer": "WIN-DC-PRIMARY", "TargetUserName": "svc-deploy", "LogonType": 5, "IpAddress": "-", "LogonProcessName": "Advapi", "EventTimeUTC": "02:50:00Z"}
{"incident_id": "INC-HUNT-002", "template_id": "hunt-dns-tunneling-large-txt", "event_index": 0, "source": "dns", "query_name": "QmFzZTY0RXh0cmFjdGVkUmVjb3JkSWQ9NTQzMjE3OTg3NjU0MzIxLnVzZXIuYWlzb2MuZGV2.exfil.45.142.122.111.in-addr.arpa", "query_type": "TXT", "query_length": 142, "src_host": "WIN-PROD-WEB02", "dst_ip": "45.142.122.111", "response_size": 512, "queries_per_hour": 187}
{"incident_id": "INC-HUNT-002", "template_id": "hunt-dns-tunneling-large-txt", "event_index": 1, "source": "dns", "query_name": "WW91U2hhbGxOb3RQYXNzVGhpc0lzREFUQVRyYW5zZmVyU3RyZWFtMQ.exfil.45.142.122.111.in-addr.arpa", "query_type": "TXT", "query_length": 121, "src_host": "WIN-PROD-WEB02", "dst_ip": "45.142.122.111", "response_size": 512, "queries_per_hour": 187}
-{"incident_id": "INC-HUNT-002-NEG", "template_id": "hunt-dns-tunneling-large-txt-neg", "event_index": 0, "source": "dns", "query_name": "_dmarc.example.com", "query_type": "TXT", "query_length": 16, "src_host": "WIN-MAIL-RELAY", "dst_ip": "8.8.8.8", "response_size": 256, "queries_per_hour": 4}
+{"incident_id": "INC-HUNT-002-NEG", "template_id": "hunt-dns-tunneling-large-txt-neg", "event_index": 0, "source": "dns", "query_name": "_dmarc.example.com", "query_type": "TXT", "query_length": 16, "src_host": "WIN-MAIL-RELAY", "dst_ip": "8.8.8.8", "response_size": 256, "queries_per_hour": 187}
{"incident_id": "INC-HUNT-003", "template_id": "hunt-oauth-app-mass-consent", "event_index": 0, "source": "m365_audit", "Workload": "AzureActiveDirectory", "Operation": "Consent to application.", "UserId": "[email protected]", "ApplicationId": "98765432-aaaa-bbbb-cccc-1234567890ab", "ApplicationName": "Mailbox Helper Pro", "ConsentedScopes": ["Mail.Read", "Mail.Send", "Files.Read.All", "offline_access"], "distinct_users_per_app_24h": 7}
{"incident_id": "INC-HUNT-003", "template_id": "hunt-oauth-app-mass-consent", "event_index": 1, "source": "m365_audit", "Workload": "AzureActiveDirectory", "Operation": "Consent to application.", "UserId": "[email protected]", "ApplicationId": "98765432-aaaa-bbbb-cccc-1234567890ab", "ApplicationName": "Mailbox Helper Pro", "ConsentedScopes": ["Mail.Read", "offline_access"], "distinct_users_per_app_24h": 7}
-{"incident_id": "INC-HUNT-003-NEG", "template_id": "hunt-oauth-app-mass-consent-neg", "event_index": 0, "source": "m365_audit", "Workload": "AzureActiveDirectory", "Operation": "Consent to application.", "UserId": "[email protected]", "ApplicationId": "11111111-2222-3333-4444-555555555555", "ApplicationName": "Microsoft Teams", "ConsentedScopes": ["User.Read"], "distinct_users_per_app_24h": 1}
+{"incident_id": "INC-HUNT-003-NEG", "template_id": "hunt-oauth-app-mass-consent-neg", "event_index": 0, "source": "m365_audit", "Workload": "AzureActiveDirectory", "Operation": "Consent to application.", "UserId": "[email protected]", "ApplicationId": "11111111-2222-3333-4444-555555555555", "ApplicationName": "Microsoft Teams", "ConsentedScopes": ["Mail.Read", "Mail.Send", "Files.Read.All", "offline_access"], "distinct_users_per_app_24h": 1}
{"incident_id": "INC-HUNT-004", "template_id": "hunt-cloud-iam-role-anomalous-assume", "event_index": 0, "source": "cloudtrail", "eventName": "AssumeRole", "eventSource": "sts.amazonaws.com", "awsRegion": "us-east-1", "userIdentity": {"type": "AssumedRole", "principalId": "AROAEXTERNALACCT123:attacker"}, "sourceIPAddress": "45.142.122.111", "requestParameters": {"roleArn": "arn:aws:iam::123456789012:role/prod-readonly", "roleSessionName": "atk-session-001"}, "principal_first_seen_against_role": true, "privileged_calls_within_5min": 12}
{"incident_id": "INC-HUNT-004", "template_id": "hunt-cloud-iam-role-anomalous-assume", "event_index": 1, "source": "cloudtrail", "eventName": "ListBuckets", "eventSource": "s3.amazonaws.com", "awsRegion": "us-east-1", "userIdentity": {"type": "AssumedRole", "principalId": "AROAEXTERNALACCT123:attacker", "sessionContext": {"sessionIssuer": {"arn": "arn:aws:iam::123456789012:role/prod-readonly"}}}, "sourceIPAddress": "45.142.122.111", "privileged_calls_within_5min": 12}
-{"incident_id": "INC-HUNT-004-NEG", "template_id": "hunt-cloud-iam-role-anomalous-assume-neg", "event_index": 0, "source": "cloudtrail", "eventName": "AssumeRole", "eventSource": "sts.amazonaws.com", "awsRegion": "us-east-1", "userIdentity": {"type": "AssumedRole", "principalId": "AROACI0123456:ci-deployer"}, "sourceIPAddress": "10.0.0.50", "requestParameters": {"roleArn": "arn:aws:iam::123456789012:role/ci-deploy", "roleSessionName": "github-action-deploy"}, "principal_first_seen_against_role": false}
+{"incident_id": "INC-HUNT-004-NEG", "template_id": "hunt-cloud-iam-role-anomalous-assume-neg", "event_index": 0, "source": "cloudtrail", "eventName": "AssumeRole", "eventSource": "sts.amazonaws.com", "awsRegion": "us-east-1", "userIdentity": {"type": "AssumedRole", "principalId": "AROACI0123456:ci-deployer"}, "sourceIPAddress": "10.0.0.50", "requestParameters": {"roleArn": "arn:aws:iam::123456789012:role/ci-deploy", "roleSessionName": "github-action-deploy"}, "principal_first_seen_against_role": false, "privileged_calls_within_5min": 12}
{"incident_id": "INC-HUNT-005", "template_id": "hunt-lolbin-rundll32-network", "event_index": 0, "source": "sysmon", "channel": "Microsoft-Windows-Sysmon/Operational", "Provider": "Microsoft-Windows-Sysmon", "EventID": 3, "Computer": "WIN-PROD-WEB02", "User": "[email protected]", "Image": "C:\\Windows\\System32\\rundll32.exe", "DestinationIp": "91.134.231.15", "DestinationPort": 443, "DestinationIp_is_public": true, "DestinationIp_is_microsoft": false}
{"incident_id": "INC-HUNT-005", "template_id": "hunt-lolbin-rundll32-network", "event_index": 1, "source": "sysmon", "channel": "Microsoft-Windows-Sysmon/Operational", "Provider": "Microsoft-Windows-Sysmon", "EventID": 1, "Computer": "WIN-PROD-WEB02", "User": "[email protected]", "Image": "C:\\Windows\\System32\\rundll32.exe", "ParentImage": "C:\\Program Files\\Microsoft Office\\root\\Office16\\WINWORD.EXE", "CommandLine": "rundll32.exe C:\\Users\\Public\\x.dll,EntryPoint"}
{"incident_id": "INC-HUNT-005-NEG", "template_id": "hunt-lolbin-rundll32-network-neg", "event_index": 0, "source": "sysmon", "channel": "Microsoft-Windows-Sysmon/Operational", "Provider": "Microsoft-Windows-Sysmon", "EventID": 3, "Computer": "WIN-PROD-WEB02", "User": "SYSTEM", "Image": "C:\\Windows\\System32\\rundll32.exe", "DestinationIp": "20.190.190.35", "DestinationPort": 443, "DestinationIp_is_public": true, "DestinationIp_is_microsoft": true}
+{"incident_id": "INC-HUNT-006", "template_id": "hunt-kerberoast-rc4-service-ticket", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4769, "TicketEncryptionType": "0x17", "ServiceName": "MSSQLSvc/db01.corp.local", "TargetUserName": "j.okafor", "IpAddress": "10.8.3.44"}
+{"incident_id": "INC-HUNT-006-NEG", "template_id": "hunt-kerberoast-rc4-service-ticket-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4769, "TicketEncryptionType": "0x12", "ServiceName": "MSSQLSvc/db01.corp.local", "TargetUserName": "j.okafor", "IpAddress": "10.8.3.44"}
+{"incident_id": "INC-HUNT-007", "template_id": "hunt-lsass-process-access", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 10, "TargetImage": "C:\\Windows\\System32\\lsass.exe", "SourceImage": "C:\\Users\\Public\\rundll32.exe", "GrantedAccess": "0x1410", "Computer": "WIN-FIN-04"}
+{"incident_id": "INC-HUNT-007-NEG", "template_id": "hunt-lsass-process-access-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 10, "TargetImage": "C:\\Windows\\System32\\lsass.exe", "SourceImage": "C:\\Program Files\\Endpoint\\agent.exe", "GrantedAccess": "0x1000", "Computer": "WIN-FIN-04"}
+{"incident_id": "INC-HUNT-008", "template_id": "hunt-dcsync-replication-rights", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4662, "Properties": "{1131f6aa-9c07-11d1-f79f-00c04fc2dcd2}", "SubjectIsDomainController": false, "SubjectUserName": "m.arnold", "Computer": "WIN-DC-PRIMARY"}
+{"incident_id": "INC-HUNT-008-NEG", "template_id": "hunt-dcsync-replication-rights-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4662, "Properties": "{1131f6aa-9c07-11d1-f79f-00c04fc2dcd2}", "SubjectIsDomainController": true, "SubjectUserName": "WIN-DC-SECONDARY$", "Computer": "WIN-DC-PRIMARY"}
+{"incident_id": "INC-HUNT-009", "template_id": "hunt-ntds-dit-shadow-copy", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "CommandLine": "vssadmin create shadow /for=C:", "IsDomainController": true, "Image": "C:\\Windows\\System32\\vssadmin.exe", "User": "CORP\\svc-backup"}
+{"incident_id": "INC-HUNT-009-NEG", "template_id": "hunt-ntds-dit-shadow-copy-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "CommandLine": "vssadmin create shadow /for=C:", "IsDomainController": false, "Image": "C:\\Windows\\System32\\vssadmin.exe", "User": "CORP\\svc-backup"}
+{"incident_id": "INC-HUNT-010", "template_id": "hunt-cloud-access-key-created-then-used-elsewhere", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "CreateAccessKey", "firstUseFromDifferentAsn": true, "userIdentity": "arn:aws:iam::1234:user/deploy", "sourceIPAddress": "203.0.113.40"}
+{"incident_id": "INC-HUNT-010-NEG", "template_id": "hunt-cloud-access-key-created-then-used-elsewhere-neg", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "CreateAccessKey", "firstUseFromDifferentAsn": false, "userIdentity": "arn:aws:iam::1234:user/deploy", "sourceIPAddress": "203.0.113.40"}
+{"incident_id": "INC-HUNT-011", "template_id": "hunt-okta-mfa-factor-reset-by-other-admin", "event_index": 0, "is_synthetic": true, "source": "okta", "eventType": "user.mfa.factor.reset_all", "actorFirstTimeResettingMfa": true, "actor": "[email protected]", "target": "[email protected]"}
+{"incident_id": "INC-HUNT-011-NEG", "template_id": "hunt-okta-mfa-factor-reset-by-other-admin-neg", "event_index": 0, "is_synthetic": true, "source": "okta", "eventType": "user.mfa.factor.reset_all", "actorFirstTimeResettingMfa": false, "actor": "[email protected]", "target": "[email protected]"}
+{"incident_id": "INC-HUNT-012", "template_id": "hunt-vault-secret-mass-read", "event_index": 0, "is_synthetic": true, "source": "vault", "operation": "read", "distinct_paths_last_hour": 214, "identity": "ci-runner-17", "namespace": "prod"}
+{"incident_id": "INC-HUNT-012-NEG", "template_id": "hunt-vault-secret-mass-read-neg", "event_index": 0, "is_synthetic": true, "source": "vault", "operation": "read", "distinct_paths_last_hour": 6, "identity": "ci-runner-17", "namespace": "prod"}
+{"incident_id": "INC-HUNT-013", "template_id": "hunt-scheduled-task-from-temp-path", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4698, "TaskCommand": "C:\\Users\\Public\\Temp\\updater.exe", "TaskName": "\\Microsoft\\Windows\\UpdateCheck", "SubjectUserName": "d.chen"}
+{"incident_id": "INC-HUNT-013-NEG", "template_id": "hunt-scheduled-task-from-temp-path-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4698, "TaskCommand": "C:\\Program Files\\Corp\\updater.exe", "TaskName": "\\Corp\\UpdateCheck", "SubjectUserName": "d.chen"}
+{"incident_id": "INC-HUNT-014", "template_id": "hunt-run-key-persistence", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 13, "TargetObject": "HKU\\S-1-5-21-9\\Software\\Microsoft\\Windows\\CurrentVersion\\Run\\OneDriveSync", "ImageIsSignedInstaller": false, "Image": "C:\\Users\\Public\\wscript.exe"}
+{"incident_id": "INC-HUNT-014-NEG", "template_id": "hunt-run-key-persistence-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 13, "TargetObject": "HKU\\S-1-5-21-9\\Software\\Microsoft\\Windows\\CurrentVersion\\Run\\OneDriveSync", "ImageIsSignedInstaller": true, "Image": "C:\\Program Files\\Microsoft OneDrive\\OneDriveSetup.exe"}
+{"incident_id": "INC-HUNT-015", "template_id": "hunt-linux-cron-added-outside-package-manager", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "syscall": "openat", "path": "/var/spool/cron/crontabs/www-data", "parentIsPackageManager": false, "comm": "sh", "auid": "33"}
+{"incident_id": "INC-HUNT-015-NEG", "template_id": "hunt-linux-cron-added-outside-package-manager-neg", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "syscall": "openat", "path": "/var/spool/cron/crontabs/www-data", "parentIsPackageManager": true, "comm": "dpkg", "auid": "0"}
+{"incident_id": "INC-HUNT-016", "template_id": "hunt-systemd-service-written-by-web-account", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "path": "/etc/systemd/system/telemetry-sync.service", "accountClass": "web", "comm": "python3", "auid": "33"}
+{"incident_id": "INC-HUNT-016-NEG", "template_id": "hunt-systemd-service-written-by-web-account-neg", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "path": "/etc/systemd/system/telemetry-sync.service", "accountClass": "admin", "comm": "ansible-playbook", "auid": "1001"}
+{"incident_id": "INC-HUNT-017", "template_id": "hunt-github-actions-workflow-added-by-outside-collaborator", "event_index": 0, "is_synthetic": true, "source": "github_audit", "action": "workflows.updated", "actorIsOrgMember": false, "actor": "drive-by-contrib", "repo": "corp/payments-api"}
+{"incident_id": "INC-HUNT-017-NEG", "template_id": "hunt-github-actions-workflow-added-by-outside-collaborator-neg", "event_index": 0, "is_synthetic": true, "source": "github_audit", "action": "workflows.updated", "actorIsOrgMember": true, "actor": "a.patel", "repo": "corp/payments-api"}
+{"incident_id": "INC-HUNT-018", "template_id": "hunt-oauth-refresh-token-long-lived-grant", "event_index": 0, "is_synthetic": true, "source": "azure_ad", "operation": "Consent to application", "appFirstSeenInTenant": true, "grantIncludesOfflineAccess": true, "appDisplayName": "Mail Analytics Pro", "user": "[email protected]"}
+{"incident_id": "INC-HUNT-018-NEG", "template_id": "hunt-oauth-refresh-token-long-lived-grant-neg", "event_index": 0, "is_synthetic": true, "source": "azure_ad", "operation": "Consent to application", "appFirstSeenInTenant": true, "grantIncludesOfflineAccess": false, "appDisplayName": "Mail Analytics Pro", "user": "[email protected]"}
+{"incident_id": "INC-HUNT-019", "template_id": "hunt-windows-event-log-cleared", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 1102, "duringMaintenanceWindow": false, "SubjectUserName": "administrator", "Computer": "WIN-FIN-04"}
+{"incident_id": "INC-HUNT-019-NEG", "template_id": "hunt-windows-event-log-cleared-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 1102, "duringMaintenanceWindow": true, "SubjectUserName": "svc-imaging", "Computer": "WIN-BUILD-01"}
+{"incident_id": "INC-HUNT-020", "template_id": "hunt-defender-exclusion-added", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 13, "TargetObject": "HKLM\\SOFTWARE\\Microsoft\\Windows Defender\\Exclusions\\Paths\\C:\\Users\\Public", "writtenByManagementAgent": false, "Image": "C:\\Windows\\System32\\reg.exe"}
+{"incident_id": "INC-HUNT-020-NEG", "template_id": "hunt-defender-exclusion-added-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 13, "TargetObject": "HKLM\\SOFTWARE\\Microsoft\\Windows Defender\\Exclusions\\Paths\\C:\\Users\\Public", "writtenByManagementAgent": true, "Image": "C:\\Program Files\\Corp MDM\\policyagent.exe"}
+{"incident_id": "INC-HUNT-021", "template_id": "hunt-timestomp-file-creation-time", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 2, "creationTimeMovedBackwardsDays": 431, "TargetFilename": "C:\\Windows\\System32\\drivers\\netio.sys", "Image": "C:\\Users\\Public\\tool.exe"}
+{"incident_id": "INC-HUNT-021-NEG", "template_id": "hunt-timestomp-file-creation-time-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 2, "creationTimeMovedBackwardsDays": 0, "TargetFilename": "C:\\Builds\\artifact.dll", "Image": "C:\\Program Files\\Corp Build\\msbuild.exe"}
+{"incident_id": "INC-HUNT-022", "template_id": "hunt-cloudtrail-logging-stopped", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "StopLogging", "performedByPipeline": false, "userIdentity": "arn:aws:iam::1234:user/contractor", "sourceIPAddress": "198.51.100.77"}
+{"incident_id": "INC-HUNT-022-NEG", "template_id": "hunt-cloudtrail-logging-stopped-neg", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "StopLogging", "performedByPipeline": true, "userIdentity": "arn:aws:iam::1234:role/terraform-apply", "sourceIPAddress": "10.0.4.9"}
+{"incident_id": "INC-HUNT-023", "template_id": "hunt-gcp-audit-sink-removed", "event_index": 0, "is_synthetic": true, "source": "gcp_audit", "methodName": "google.logging.v2.ConfigServiceV2.DeleteSink", "performedByPipeline": false, "principalEmail": "[email protected]", "resource": "projects/prod-data"}
+{"incident_id": "INC-HUNT-023-NEG", "template_id": "hunt-gcp-audit-sink-removed-neg", "event_index": 0, "is_synthetic": true, "source": "gcp_audit", "methodName": "google.logging.v2.ConfigServiceV2.DeleteSink", "performedByPipeline": true, "principalEmail": "[email protected]", "resource": "projects/prod-data"}
+{"incident_id": "INC-HUNT-024", "template_id": "hunt-k8s-audit-policy-weakened", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "verb": "delete", "resource": "validatingwebhookconfigurations", "performedByPipeline": false, "user": "[email protected]", "namespace": "kube-system"}
+{"incident_id": "INC-HUNT-024-NEG", "template_id": "hunt-k8s-audit-policy-weakened-neg", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "verb": "delete", "resource": "validatingwebhookconfigurations", "performedByPipeline": true, "user": "system:serviceaccount:flux:reconciler", "namespace": "kube-system"}
+{"incident_id": "INC-HUNT-025", "template_id": "hunt-office-child-process-script-host", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "ParentImage": "C:\\Program Files\\Microsoft Office\\root\\Office16\\WINWORD.EXE", "Image": "C:\\Windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe", "CommandLine": "powershell -nop -w hidden -enc SQBFAFgA", "User": "CORP\\l.dubois"}
+{"incident_id": "INC-HUNT-025-NEG", "template_id": "hunt-office-child-process-script-host-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "ParentImage": "C:\\Program Files\\Microsoft Office\\root\\Office16\\WINWORD.EXE", "Image": "C:\\Windows\\System32\\splwow64.exe", "CommandLine": "splwow64.exe 12288", "User": "CORP\\l.dubois"}
+{"incident_id": "INC-HUNT-026", "template_id": "hunt-encoded-powershell-command", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "CommandLine": "powershell.exe -nop -enc SQBuAHYAbwBrAGUALQBXAGUAYgBSAGUAcQB1AGUAcwB0AA==", "userIsAdministrator": false, "Image": "C:\\Windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe", "User": "CORP\\t.nakamura"}
+{"incident_id": "INC-HUNT-026-NEG", "template_id": "hunt-encoded-powershell-command-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "CommandLine": "powershell.exe -nop -enc SQBuAHYAbwBrAGUALQBXAGUAYgBSAGUAcQB1AGUAcwB0AA==", "userIsAdministrator": true, "Image": "C:\\Windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe", "User": "CORP\\svc-sccm"}
+{"incident_id": "INC-HUNT-027", "template_id": "hunt-container-exec-into-production-pod", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "subresource": "exec", "namespaceClass": "production", "user": "[email protected]", "pod": "payments-api-7d9f", "verb": "create"}
+{"incident_id": "INC-HUNT-027-NEG", "template_id": "hunt-container-exec-into-production-pod-neg", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "subresource": "exec", "namespaceClass": "development", "user": "[email protected]", "pod": "payments-api-dev-2a1c", "verb": "create"}
+{"incident_id": "INC-HUNT-028", "template_id": "hunt-wmi-remote-process-creation", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "ParentImage": "C:\\Windows\\System32\\wbem\\WmiPrvSE.exe", "hostClass": "workstation", "Image": "C:\\Windows\\System32\\cmd.exe", "CommandLine": "cmd.exe /c whoami"}
+{"incident_id": "INC-HUNT-028-NEG", "template_id": "hunt-wmi-remote-process-creation-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "ParentImage": "C:\\Windows\\System32\\wbem\\WmiPrvSE.exe", "hostClass": "management-server", "Image": "C:\\Windows\\System32\\cmd.exe", "CommandLine": "cmd.exe /c whoami"}
+{"incident_id": "INC-HUNT-029", "template_id": "hunt-ad-enumeration-burst", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4661, "objectsEnumeratedLast10Min": 3120, "hostClass": "workstation", "SubjectUserName": "p.moreau", "Computer": "WIN-MKT-22"}
+{"incident_id": "INC-HUNT-029-NEG", "template_id": "hunt-ad-enumeration-burst-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4661, "objectsEnumeratedLast10Min": 3120, "hostClass": "identity-management-server", "SubjectUserName": "svc-idm", "Computer": "WIN-IDM-01"}
+{"incident_id": "INC-HUNT-030", "template_id": "hunt-cloud-permission-enumeration", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "ListBuckets", "distinctServicesLast15Min": 27, "performedByPipeline": false, "userIdentity": "arn:aws:iam::1234:user/contractor", "sourceIPAddress": "198.51.100.77"}
+{"incident_id": "INC-HUNT-030-NEG", "template_id": "hunt-cloud-permission-enumeration-neg", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "ListBuckets", "distinctServicesLast15Min": 27, "performedByPipeline": true, "userIdentity": "arn:aws:iam::1234:role/cspm-scanner", "sourceIPAddress": "10.0.4.9"}
+{"incident_id": "INC-HUNT-031", "template_id": "hunt-network-share-enumeration", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 5140, "ShareName": "\\\\*\\ADMIN$", "distinctPeersLast10Min": 64, "SubjectUserName": "p.moreau", "IpAddress": "10.4.19.88"}
+{"incident_id": "INC-HUNT-031-NEG", "template_id": "hunt-network-share-enumeration-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 5140, "ShareName": "\\\\*\\ADMIN$", "distinctPeersLast10Min": 2, "SubjectUserName": "svc-backup", "IpAddress": "10.4.2.10"}
+{"incident_id": "INC-HUNT-032", "template_id": "hunt-smb-admin-share-write-executable", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 5145, "RelativeTargetName": "windows\\temp\\svc.exe", "AccessMask": "0x2", "ShareName": "\\\\*\\ADMIN$", "SubjectUserName": "p.moreau"}
+{"incident_id": "INC-HUNT-032-NEG", "template_id": "hunt-smb-admin-share-write-executable-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 5145, "RelativeTargetName": "windows\\temp\\svc.exe", "AccessMask": "0x1", "ShareName": "\\\\*\\ADMIN$", "SubjectUserName": "svc-sccm"}
+{"incident_id": "INC-HUNT-033", "template_id": "hunt-rdp-from-workstation-to-workstation", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4624, "LogonType": 10, "sourceHostClass": "workstation", "targetHostClass": "workstation", "TargetUserName": "p.moreau", "Computer": "WIN-MKT-31"}
+{"incident_id": "INC-HUNT-033-NEG", "template_id": "hunt-rdp-from-workstation-to-workstation-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4624, "LogonType": 10, "sourceHostClass": "jump-host", "targetHostClass": "workstation", "TargetUserName": "helpdesk-02", "Computer": "WIN-MKT-31"}
+{"incident_id": "INC-HUNT-034", "template_id": "hunt-ssh-key-added-to-authorized-keys", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "path": "/home/deploy/.ssh/authorized_keys", "parentIsConfigManagement": false, "comm": "bash", "auid": "1042"}
+{"incident_id": "INC-HUNT-034-NEG", "template_id": "hunt-ssh-key-added-to-authorized-keys-neg", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "path": "/home/deploy/.ssh/authorized_keys", "parentIsConfigManagement": true, "comm": "ansible-playbook", "auid": "1001"}
+{"incident_id": "INC-HUNT-035", "template_id": "hunt-tailscale-device-key-expiry-disabled", "event_index": 0, "is_synthetic": true, "source": "tailscale", "action": "device.keyExpiry.disable", "actorIsTailnetAdmin": false, "actor": "[email protected]", "device": "laptop-contractor-3"}
+{"incident_id": "INC-HUNT-035-NEG", "template_id": "hunt-tailscale-device-key-expiry-disabled-neg", "event_index": 0, "is_synthetic": true, "source": "tailscale", "action": "device.keyExpiry.disable", "actorIsTailnetAdmin": true, "actor": "[email protected]", "device": "gw-eu-west-1"}
+{"incident_id": "INC-HUNT-036", "template_id": "hunt-saas-bulk-download", "event_index": 0, "is_synthetic": true, "source": "m365_audit", "Operation": "FileDownloaded", "filesDownloadedLastHour": 1840, "UserId": "[email protected]", "SiteUrl": "/sites/Finance"}
+{"incident_id": "INC-HUNT-036-NEG", "template_id": "hunt-saas-bulk-download-neg", "event_index": 0, "is_synthetic": true, "source": "m365_audit", "Operation": "FileDownloaded", "filesDownloadedLastHour": 11, "UserId": "[email protected]", "SiteUrl": "/sites/Finance"}
+{"incident_id": "INC-HUNT-037", "template_id": "hunt-snowflake-large-result-export", "event_index": 0, "is_synthetic": true, "source": "snowflake", "query_type": "UNLOAD", "rows_unloaded": 42000000, "performedByPipeline": false, "user_name": "ANALYST_RS", "warehouse": "ADHOC_WH"}
+{"incident_id": "INC-HUNT-037-NEG", "template_id": "hunt-snowflake-large-result-export-neg", "event_index": 0, "is_synthetic": true, "source": "snowflake", "query_type": "UNLOAD", "rows_unloaded": 42000000, "performedByPipeline": true, "user_name": "ETL_SERVICE", "warehouse": "LOAD_WH"}
+{"incident_id": "INC-HUNT-038", "template_id": "hunt-salesforce-report-export-by-departing-user", "event_index": 0, "is_synthetic": true, "source": "salesforce", "EventType": "ReportExport", "rowsExported": 24500, "userIsInNoticePeriod": true, "UserId": "005xx00000ABcDe", "ReportName": "All Accounts and Contacts"}
+{"incident_id": "INC-HUNT-038-NEG", "template_id": "hunt-salesforce-report-export-by-departing-user-neg", "event_index": 0, "is_synthetic": true, "source": "salesforce", "EventType": "ReportExport", "rowsExported": 24500, "userIsInNoticePeriod": false, "UserId": "005xx00000ZyXwV", "ReportName": "All Accounts and Contacts"}
+{"incident_id": "INC-HUNT-039", "template_id": "hunt-archive-created-then-uploaded", "event_index": 0, "is_synthetic": true, "source": "proxy", "http_method": "POST", "destination_category": "personal-storage", "upload_bytes": 1240000000, "user": "j.okafor", "url_host": "personal-drive.example"}
+{"incident_id": "INC-HUNT-039-NEG", "template_id": "hunt-archive-created-then-uploaded-neg", "event_index": 0, "is_synthetic": true, "source": "proxy", "http_method": "POST", "destination_category": "corporate-storage", "upload_bytes": 1240000000, "user": "j.okafor", "url_host": "corp-sharepoint.example"}
+{"incident_id": "INC-HUNT-040", "template_id": "hunt-email-forwarding-rule-external", "event_index": 0, "is_synthetic": true, "source": "m365_audit", "Operation": "New-InboxRule", "forwardsExternally": true, "UserId": "[email protected]", "ForwardTo": "[email protected]"}
+{"incident_id": "INC-HUNT-040-NEG", "template_id": "hunt-email-forwarding-rule-external-neg", "event_index": 0, "is_synthetic": true, "source": "m365_audit", "Operation": "New-InboxRule", "forwardsExternally": false, "UserId": "[email protected]", "ForwardTo": "[email protected]"}
+{"incident_id": "INC-HUNT-041", "template_id": "hunt-beacon-regular-interval", "event_index": 0, "is_synthetic": true, "source": "proxy", "connection_count_last_6h": 712, "interval_jitter_ratio": 0.01, "destination_is_known_service": false, "src_host": "WIN-FIN-04", "url_host": "cdn-metrics.example"}
+{"incident_id": "INC-HUNT-041-NEG", "template_id": "hunt-beacon-regular-interval-neg", "event_index": 0, "is_synthetic": true, "source": "proxy", "connection_count_last_6h": 712, "interval_jitter_ratio": 0.01, "destination_is_known_service": true, "src_host": "WIN-FIN-04", "url_host": "telemetry.microsoft.com"}
+{"incident_id": "INC-HUNT-042", "template_id": "hunt-dns-over-https-to-unapproved-resolver", "event_index": 0, "is_synthetic": true, "source": "proxy", "url_path": "/dns-query", "resolverIsApproved": false, "src_host": "WIN-FIN-04", "url_host": "doh.unknown.example"}
+{"incident_id": "INC-HUNT-042-NEG", "template_id": "hunt-dns-over-https-to-unapproved-resolver-neg", "event_index": 0, "is_synthetic": true, "source": "proxy", "url_path": "/dns-query", "resolverIsApproved": true, "src_host": "WIN-FIN-04", "url_host": "doh.corp.example"}
+{"incident_id": "INC-HUNT-043", "template_id": "hunt-newly-registered-domain-contact", "event_index": 0, "is_synthetic": true, "source": "proxy", "domain_age_days": 2, "destination_category": "uncategorised", "src_host": "WIN-MKT-22", "url_host": "secure-invoice-portal.example"}
+{"incident_id": "INC-HUNT-043-NEG", "template_id": "hunt-newly-registered-domain-contact-neg", "event_index": 0, "is_synthetic": true, "source": "proxy", "domain_age_days": 2, "destination_category": "business-software", "src_host": "WIN-MKT-22", "url_host": "new-vendor-portal.example"}
+{"incident_id": "INC-HUNT-044", "template_id": "hunt-tor-exit-node-connection", "event_index": 0, "is_synthetic": true, "source": "firewall", "action": "allow", "destination_is_anonymising_relay": true, "src_ip": "10.4.19.88", "dst_ip": "198.51.100.211", "dst_port": 9001}
+{"incident_id": "INC-HUNT-044-NEG", "template_id": "hunt-tor-exit-node-connection-neg", "event_index": 0, "is_synthetic": true, "source": "firewall", "action": "deny", "destination_is_anonymising_relay": true, "src_ip": "10.4.19.88", "dst_ip": "198.51.100.211", "dst_port": 9001}
+{"incident_id": "INC-HUNT-045", "template_id": "hunt-shadow-copy-deletion", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "CommandLine": "vssadmin.exe delete shadows /all /quiet", "Image": "C:\\Windows\\System32\\vssadmin.exe", "User": "CORP\\p.moreau"}
+{"incident_id": "INC-HUNT-045-NEG", "template_id": "hunt-shadow-copy-deletion-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 1, "CommandLine": "vssadmin.exe list shadows", "Image": "C:\\Windows\\System32\\vssadmin.exe", "User": "CORP\\svc-backup"}
+{"incident_id": "INC-HUNT-046", "template_id": "hunt-mass-file-rename-extension", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 11, "filesRenamedLast5Min": 8412, "extensionIsKnownApplication": false, "TargetFilename": "D:\\Shares\\Finance\\ledger.xlsx.a7z9", "Image": "C:\\Users\\Public\\svc.exe"}
+{"incident_id": "INC-HUNT-046-NEG", "template_id": "hunt-mass-file-rename-extension-neg", "event_index": 0, "is_synthetic": true, "source": "sysmon", "EventID": 11, "filesRenamedLast5Min": 8412, "extensionIsKnownApplication": true, "TargetFilename": "D:\\Media\\render_0412.mov", "Image": "C:\\Program Files\\Render\\encoder.exe"}
+{"incident_id": "INC-HUNT-047", "template_id": "hunt-s3-bucket-made-public", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "PutBucketPolicy", "resultsInPublicAccess": true, "bucket": "corp-customer-exports", "userIdentity": "arn:aws:iam::1234:user/contractor"}
+{"incident_id": "INC-HUNT-047-NEG", "template_id": "hunt-s3-bucket-made-public-neg", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "PutBucketPolicy", "resultsInPublicAccess": false, "bucket": "corp-customer-exports", "userIdentity": "arn:aws:iam::1234:role/terraform-apply"}
+{"incident_id": "INC-HUNT-048", "template_id": "hunt-backup-job-disabled", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4701, "TaskName": "\\Corp\\NightlyBackup", "duringChangeWindow": false, "SubjectUserName": "p.moreau", "Computer": "WIN-FS-01"}
+{"incident_id": "INC-HUNT-048-NEG", "template_id": "hunt-backup-job-disabled-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4701, "TaskName": "\\Corp\\NightlyBackup", "duringChangeWindow": true, "SubjectUserName": "svc-backup-admin", "Computer": "WIN-FS-01"}
+{"incident_id": "INC-HUNT-049", "template_id": "hunt-privileged-group-membership-added", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4728, "groupIsPrivileged": true, "viaApprovedProvisioning": false, "TargetUserName": "p.moreau", "GroupName": "Domain Admins"}
+{"incident_id": "INC-HUNT-049-NEG", "template_id": "hunt-privileged-group-membership-added-neg", "event_index": 0, "is_synthetic": true, "source": "windows_security", "EventID": 4728, "groupIsPrivileged": true, "viaApprovedProvisioning": true, "TargetUserName": "new.admin", "GroupName": "Domain Admins"}
+{"incident_id": "INC-HUNT-050", "template_id": "hunt-aws-role-trust-policy-widened", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "UpdateAssumeRolePolicy", "trustsExternalPrincipal": true, "roleName": "prod-data-reader", "userIdentity": "arn:aws:iam::1234:user/contractor"}
+{"incident_id": "INC-HUNT-050-NEG", "template_id": "hunt-aws-role-trust-policy-widened-neg", "event_index": 0, "is_synthetic": true, "source": "aws_cloudtrail", "eventName": "UpdateAssumeRolePolicy", "trustsExternalPrincipal": false, "roleName": "prod-data-reader", "userIdentity": "arn:aws:iam::1234:role/terraform-apply"}
+{"incident_id": "INC-HUNT-051", "template_id": "hunt-k8s-cluster-admin-binding", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "verb": "create", "roleRefName": "cluster-admin", "subjectKind": "ServiceAccount", "subjectName": "build-runner", "namespace": "ci"}
+{"incident_id": "INC-HUNT-051-NEG", "template_id": "hunt-k8s-cluster-admin-binding-neg", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "verb": "create", "roleRefName": "cluster-admin", "subjectKind": "User", "subjectName": "[email protected]", "namespace": "kube-system"}
+{"incident_id": "INC-HUNT-052", "template_id": "hunt-sudoers-modified-outside-config-management", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "path": "/etc/sudoers.d/90-deploy", "parentIsConfigManagement": false, "comm": "vi", "auid": "1042"}
+{"incident_id": "INC-HUNT-052-NEG", "template_id": "hunt-sudoers-modified-outside-config-management-neg", "event_index": 0, "is_synthetic": true, "source": "linux_auditd", "path": "/etc/sudoers.d/90-deploy", "parentIsConfigManagement": true, "comm": "ansible-playbook", "auid": "1001"}
+{"incident_id": "INC-HUNT-053", "template_id": "hunt-gcp-service-account-key-created", "event_index": 0, "is_synthetic": true, "source": "gcp_audit", "methodName": "google.iam.admin.v1.CreateServiceAccountKey", "performedByPipeline": false, "principalEmail": "[email protected]", "resource": "projects/prod-data"}
+{"incident_id": "INC-HUNT-053-NEG", "template_id": "hunt-gcp-service-account-key-created-neg", "event_index": 0, "is_synthetic": true, "source": "gcp_audit", "methodName": "google.iam.admin.v1.CreateServiceAccountKey", "performedByPipeline": true, "principalEmail": "[email protected]", "resource": "projects/prod-data"}
+{"incident_id": "INC-HUNT-054", "template_id": "hunt-impossible-travel-sign-in", "event_index": 0, "is_synthetic": true, "source": "okta", "outcome": "SUCCESS", "impliedTravelSpeedKph": 4200, "viaCorporateVpn": false, "actor": "[email protected]", "clientCountry": "SG"}
+{"incident_id": "INC-HUNT-054-NEG", "template_id": "hunt-impossible-travel-sign-in-neg", "event_index": 0, "is_synthetic": true, "source": "okta", "outcome": "SUCCESS", "impliedTravelSpeedKph": 4200, "viaCorporateVpn": true, "actor": "[email protected]", "clientCountry": "SG"}
+{"incident_id": "INC-HUNT-055", "template_id": "hunt-password-spray-low-and-slow", "event_index": 0, "is_synthetic": true, "source": "okta", "outcome": "FAILURE", "distinctAccountsFromIpLastHour": 214, "attemptsPerAccount": 2, "clientIp": "198.51.100.77"}
+{"incident_id": "INC-HUNT-055-NEG", "template_id": "hunt-password-spray-low-and-slow-neg", "event_index": 0, "is_synthetic": true, "source": "okta", "outcome": "FAILURE", "distinctAccountsFromIpLastHour": 1, "attemptsPerAccount": 2, "clientIp": "10.4.12.55"}
+{"incident_id": "INC-HUNT-056", "template_id": "hunt-vpn-login-from-hosting-provider", "event_index": 0, "is_synthetic": true, "source": "vpn", "status": "connected", "sourceAsnClass": "hosting", "user": "t.nakamura", "src_ip": "198.51.100.140"}
+{"incident_id": "INC-HUNT-056-NEG", "template_id": "hunt-vpn-login-from-hosting-provider-neg", "event_index": 0, "is_synthetic": true, "source": "vpn", "status": "connected", "sourceAsnClass": "residential", "user": "t.nakamura", "src_ip": "203.0.113.12"}
+{"incident_id": "INC-HUNT-057", "template_id": "hunt-jamf-device-unenrolled", "event_index": 0, "is_synthetic": true, "source": "jamf", "event": "ComputerRemoved", "actorIsItAdmin": false, "actor": "t.nakamura", "device": "MBP-ENG-118"}
+{"incident_id": "INC-HUNT-057-NEG", "template_id": "hunt-jamf-device-unenrolled-neg", "event_index": 0, "is_synthetic": true, "source": "jamf", "event": "ComputerRemoved", "actorIsItAdmin": true, "actor": "[email protected]", "device": "MBP-ENG-118"}
+{"incident_id": "INC-HUNT-058", "template_id": "hunt-netskope-unsanctioned-app-upload", "event_index": 0, "is_synthetic": true, "source": "netskope", "activity": "Upload", "appSanctionState": "unsanctioned", "upload_bytes": 340000000, "user": "[email protected]", "app": "FileShare Anywhere"}
+{"incident_id": "INC-HUNT-058-NEG", "template_id": "hunt-netskope-unsanctioned-app-upload-neg", "event_index": 0, "is_synthetic": true, "source": "netskope", "activity": "Upload", "appSanctionState": "sanctioned", "upload_bytes": 340000000, "user": "[email protected]", "app": "Corporate Drive"}
+{"incident_id": "INC-HUNT-059", "template_id": "hunt-duo-bypass-code-generated", "event_index": 0, "is_synthetic": true, "source": "duo", "action": "bypass_create", "hasOpenLockoutTicket": false, "actor": "[email protected]", "target": "[email protected]"}
+{"incident_id": "INC-HUNT-059-NEG", "template_id": "hunt-duo-bypass-code-generated-neg", "event_index": 0, "is_synthetic": true, "source": "duo", "action": "bypass_create", "hasOpenLockoutTicket": true, "actor": "[email protected]", "target": "[email protected]"}
+{"incident_id": "INC-HUNT-060", "template_id": "hunt-slack-external-shared-channel-with-finance", "event_index": 0, "is_synthetic": true, "source": "slack_audit", "action": "external_shared_channel_connected", "channelSensitivity": "restricted", "channel": "finance-close", "external_team": "vendor-partners"}
+{"incident_id": "INC-HUNT-060-NEG", "template_id": "hunt-slack-external-shared-channel-with-finance-neg", "event_index": 0, "is_synthetic": true, "source": "slack_audit", "action": "external_shared_channel_connected", "channelSensitivity": "general", "channel": "vendor-support", "external_team": "vendor-partners"}
+{"incident_id": "INC-HUNT-061", "template_id": "hunt-gitlab-protected-branch-rule-removed", "event_index": 0, "is_synthetic": true, "source": "gitlab_audit", "action": "protected_branch_removed", "repoDeploysToProduction": true, "actor": "[email protected]", "repo": "corp/payments-api"}
+{"incident_id": "INC-HUNT-061-NEG", "template_id": "hunt-gitlab-protected-branch-rule-removed-neg", "event_index": 0, "is_synthetic": true, "source": "gitlab_audit", "action": "protected_branch_removed", "repoDeploysToProduction": false, "actor": "[email protected]", "repo": "corp/design-sandbox"}
+{"incident_id": "INC-HUNT-062", "template_id": "hunt-osquery-agent-stopped-reporting", "event_index": 0, "is_synthetic": true, "source": "osquery", "agentSilentMinutes": 340, "hostStillReachable": true, "host_identifier": "WIN-FIN-04"}
+{"incident_id": "INC-HUNT-062-NEG", "template_id": "hunt-osquery-agent-stopped-reporting-neg", "event_index": 0, "is_synthetic": true, "source": "osquery", "agentSilentMinutes": 340, "hostStillReachable": false, "host_identifier": "MBP-ENG-118"}
+{"incident_id": "INC-HUNT-063", "template_id": "hunt-zoom-recording-downloaded-externally", "event_index": 0, "is_synthetic": true, "source": "zoom", "event": "recording.downloaded", "downloaderIsInternal": false, "downloader": "[email protected]", "topic": "Board pre-read"}
+{"incident_id": "INC-HUNT-063-NEG", "template_id": "hunt-zoom-recording-downloaded-externally-neg", "event_index": 0, "is_synthetic": true, "source": "zoom", "event": "recording.downloaded", "downloaderIsInternal": true, "downloader": "[email protected]", "topic": "Board pre-read"}
+{"incident_id": "INC-HUNT-064", "template_id": "hunt-docker-privileged-container-launched", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "verb": "create", "resource": "pods", "podIsPrivileged": true, "namespaceClass": "production", "user": "[email protected]", "pod": "debug-shell"}
+{"incident_id": "INC-HUNT-064-NEG", "template_id": "hunt-docker-privileged-container-launched-neg", "event_index": 0, "is_synthetic": true, "source": "kubernetes_audit", "verb": "create", "resource": "pods", "podIsPrivileged": false, "namespaceClass": "production", "user": "system:serviceaccount:flux:reconciler", "pod": "payments-api-7d9f"}
+{"incident_id": "INC-HUNT-065", "template_id": "hunt-ai-gateway-prompt-with-credential-shape", "event_index": 0, "is_synthetic": true, "source": "ai_gateway", "promptContainsSecretShape": true, "destinationIsSelfHosted": false, "user": "[email protected]", "model": "external-provider/large"}
+{"incident_id": "INC-HUNT-065-NEG", "template_id": "hunt-ai-gateway-prompt-with-credential-shape-neg", "event_index": 0, "is_synthetic": true, "source": "ai_gateway", "promptContainsSecretShape": true, "destinationIsSelfHosted": true, "user": "[email protected]", "model": "self-hosted/small"}
+{"incident_id": "INC-HUNT-066", "template_id": "hunt-ai-gateway-spend-spike", "event_index": 0, "is_synthetic": true, "source": "ai_gateway", "spendRatioToBaseline": 148, "requestsLastHour": 9400, "api_key_id": "key-7741", "user": "ci-runner-17"}
+{"incident_id": "INC-HUNT-066-NEG", "template_id": "hunt-ai-gateway-spend-spike-neg", "event_index": 0, "is_synthetic": true, "source": "ai_gateway", "spendRatioToBaseline": 1.4, "requestsLastHour": 9400, "api_key_id": "key-7741", "user": "ci-runner-17"}
+{"incident_id": "INC-HUNT-067", "template_id": "hunt-imperva-waf-blocked-then-allowed-same-path", "event_index": 0, "is_synthetic": true, "source": "imperva", "priorBlocksSamePath": 143, "action": "allow", "client_ip": "198.51.100.77", "url_path": "/api/v1/export"}
+{"incident_id": "INC-HUNT-067-NEG", "template_id": "hunt-imperva-waf-blocked-then-allowed-same-path-neg", "event_index": 0, "is_synthetic": true, "source": "imperva", "priorBlocksSamePath": 143, "action": "block", "client_ip": "198.51.100.77", "url_path": "/api/v1/export"}
+{"incident_id": "INC-HUNT-068", "template_id": "hunt-confluence-space-permissions-opened", "event_index": 0, "is_synthetic": true, "source": "confluence_audit", "action": "space_permissions_updated", "grantsAllAuthenticated": true, "spaceSensitivity": "restricted", "space": "SECOPS", "actor": "[email protected]"}
+{"incident_id": "INC-HUNT-068-NEG", "template_id": "hunt-confluence-space-permissions-opened-neg", "event_index": 0, "is_synthetic": true, "source": "confluence_audit", "action": "space_permissions_updated", "grantsAllAuthenticated": false, "spaceSensitivity": "restricted", "space": "SECOPS", "actor": "[email protected]"}