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.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
chore(release): v8.1.0 (#724)
Beenu Aroracd19af54f38f
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 169dc99bf..ff000e05c 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -7,6 +7,39 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0## [Unreleased]+## [8.1.0] — 2026-09-23++**Wave-2 features, and the gaps behind them.** Every item in the wave-2+backlog ([#362](https://github.com/beenuar/AiSOC/issues/362)) was audited+against the tree before any code was written, and the backlog turned out to be+wrong in both directions: two items were already built, and four had the+capability present with the path that feeds it broken. That is the same shape+v8.0 found a dozen times — the mechanism exists, is unit-tested, and has no+caller on the path that needs it — so auditing first is now the opening step+of a wave rather than an optional one.++Three of those gaps had security consequences, and one is worth stating+plainly because it inverts what the feature appeared to do: **a Slack or Teams+approval authorized nobody.** The bots verified who clicked, recorded them in+an audit event, and called the actions service with no approver, so the+permission-tier check and separation of duties were both skipped. An approval+path that does not authorize is worse than none, because it reads as a+control.++:::warning Breaking change for ChatOps approvals+`AISOC_ACTIONS_REQUIRE_APPROVER` defaults to `true`. If you use Slack or Teams+approvals you **must** populate `AISOC_CHATOPS_APPROVERS` or approvals will be+refused with a 403. Setup: `apps/docs/docs/operations/action-approvals.md`.+:::++Also in this release: the Codespaces quickstart can start Docker for the first+time, service images are published for arm64 so Apple Silicon can run+`pnpm aisoc:demo` at all, and packaging moves to v8.2 — the blocker is+registry credentials rather than code, and a release cannot schedule an+account action by writing a version number.++Claim-to-gate matrix: **108 rows — 99 GATED, 9 PARTIAL, 0 NO GATE**.+### Added- **Securing the customer's AI estate.** AiSOC could already ingest an@@ -151,6 +184,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0### Fixed+- **The sandbox determinism test compared a timer.** Its `VOLATILE` list named+ four fields the CLI does not emit — the real one is `elapsed_ms` — so the+ comparison included a wall-clock value and failed whenever two runs+ straddled a millisecond boundary, reporting "Something in the reasoner+ depends on salted hashing again" and sending a reader after a `hash()` call+ that was not there. Volatile fields are matched by suffix now, because names+ are the thing that drifts and the `_ms` convention is not.+- **The devcontainer cold-start gate could only ever validate the previous+ image.** It probed the published `:latest` even on a pull request, by design+ — so a Dockerfile change was unverifiable until after it shipped, which is+ how a devcontainer whose non-root user was not in the `docker` group reached+ `main`. A PR touching `.devcontainer/**` now builds from its own source and+ probes that, and the trigger includes `.devcontainer/**` at all (it fired+ only on changes to the workflow file).- **The Codespaces quickstart could never start Docker([#716](https://github.com/beenuar/AiSOC/issues/716)).** The READMEadvertises Codespaces as "the zero-install way to drive the real stack in adiff --git a/README.md b/README.mdindex eb8d2d9eb..0496729b3 100644--- a/README.md+++ b/README.md@@ -7,7 +7,7 @@An open-source, self-hostable AI SOC. The agent's prompts, tool calls, and rationale are logged step-by-step and replayable. MIT-licensed.[](https://opensource.org/licenses/MIT)-[](CHANGELOG.md)+[](CHANGELOG.md)[](https://github.com/beenuar/AiSOC/actions/workflows/ci.yml)[](https://github.com/beenuar/AiSOC/actions/workflows/codeql.yml)[](https://securityscorecards.dev/viewer/?uri=github.com/beenuar/AiSOC)@@ -172,7 +172,7 @@ Full architecture (every service, every storage role, the v1.5 console workbenchA handful of headline capabilities — the rest are catalogued in [`apps/docs/docs/features/`](apps/docs/docs/features/) and indexed at the top of [`apps/docs/docs/intro.md`](apps/docs/docs/intro.md):-> **Maturity (v8.0.0 — Close the loop).** The end-to-end spine is wired and CI-gated: ingest → ClickHouse lake → live detection → fused alert → auto-triage → governed response. v8.0 connected capabilities the codebase already contained but never called. Response actions are now **verified against the vendor** rather than assumed from a 200 — an action that reports success while the effect is absent is reported as failed, because that gap is how a SOC comes to believe a host is contained when it is not. Autonomy is governed by **each tenant's own** L0–L4 policy instead of one deployment-wide environment variable. Repeat-alert suppression, shipped in v7.7, **could never fire** — the evidence fingerprint included the alert id, so no repeat ever matched; it now does. Analysts can **hunt the events AiSOC itself ingested** instead of needing an external SIEM. A verdict citing indicators the evidence never contained is demoted to human review rather than auto-closed. Autonomous *response* still defaults to copilot/dry-run: an autonomy policy governs every real execution. The live-agent LLM benchmark is preview (the deterministic-tier scoreboard is CI-gated per PR); substrate eval suites are GA. Every product claim is backed by a failing test — [claim-to-gate matrix](docs/audit/CLAIM_TO_GATE_MATRIX.md): 99 GATED / 9 PARTIAL / 0 NO GATE. Full per-claim status: [`docs/audit/REALITY_REPORT.md`](docs/audit/REALITY_REPORT.md).+> **Maturity (v8.1.0 — Wave-2 features, and the gaps behind them).** Every wave-2 backlog item was audited against the tree before any code was written, and the backlog was wrong in both directions: two were already built, four had the capability present with the path that feeds it broken. One inverted what the feature appeared to do — a Slack or Teams **approval authorized nobody**: the bots verified who clicked, logged them, and called the actions service with no approver, so the permission tier and separation of duties were both skipped. The end-to-end spine is wired and CI-gated: ingest → ClickHouse lake → live detection → fused alert → auto-triage → governed response. v8.0 connected capabilities the codebase already contained but never called. Response actions are now **verified against the vendor** rather than assumed from a 200 — an action that reports success while the effect is absent is reported as failed, because that gap is how a SOC comes to believe a host is contained when it is not. Autonomy is governed by **each tenant's own** L0–L4 policy instead of one deployment-wide environment variable. Repeat-alert suppression, shipped in v7.7, **could never fire** — the evidence fingerprint included the alert id, so no repeat ever matched; it now does. Analysts can **hunt the events AiSOC itself ingested** instead of needing an external SIEM. A verdict citing indicators the evidence never contained is demoted to human review rather than auto-closed. Autonomous *response* still defaults to copilot/dry-run: an autonomy policy governs every real execution. The live-agent LLM benchmark is preview (the deterministic-tier scoreboard is CI-gated per PR); substrate eval suites are GA. Every product claim is backed by a failing test — [claim-to-gate matrix](docs/audit/CLAIM_TO_GATE_MATRIX.md): 99 GATED / 9 PARTIAL / 0 NO GATE. Full per-claim status: [`docs/audit/REALITY_REPORT.md`](docs/audit/REALITY_REPORT.md).- **84 click-and-connect data connectors** (EDR/XDR, SIEM, NDR, cloud, CNAPP, identity, SaaS, VCS, K8s audit, network) with schema-driven config, live `Test connection`, and vault-encrypted secrets — recently adding Qualys, GreyNoise, JumpCloud, Darktrace, and Imperva alongside IBM QRadar, Netskope, Zeek/Suricata NDR, and more. One query runs SIEM-agnostic **federated search** across Splunk SPL / Sentinel KQL / Elastic ES|QL / **QRadar AQL**. Walkthrough: [`apps/docs/docs/connectors/index.md`](apps/docs/docs/connectors/index.md).- **End-to-end SIEM spine** — a cold `docker compose up` ingests connector data → lands it in the ClickHouse event lake → the executable detection corpus (833 rules) fires on the live stream → a fused alert is created, all asserted by an extended integration gate. Fuse-time threat-intel + CISA-KEV enrichment now feeds the confidence score and exploit-in-wild boost, and **stateful/windowed detections** (brute-force, password-spray, port-scan) run alongside the corpus. [`apps/docs/docs/architecture.md`](apps/docs/docs/architecture.md).diff --git a/RELEASES.md b/RELEASES.mdindex b309fef17..5497cab1f 100644--- a/RELEASES.md+++ b/RELEASES.md@@ -2,13 +2,31 @@This file mirrors what used to live in the "What's new" section of [`README.md`](README.md). The complete, machine-readable inventory (with file paths, env-var diffs, and per-release test counts) lives in [`CHANGELOG.md`](CHANGELOG.md).-> **TL;DR for first-time visitors:** AiSOC is on `v8.0.0`, released 2026-09-22 — the **Close the loop** release. It connected capabilities the codebase already contained but never called: response actions are verified against the vendor rather than assumed from a 200, autonomy is governed by each tenant's own L0–L4 policy instead of one deployment-wide environment variable, repeat-alert suppression can fire at all, and analysts can hunt the events AiSOC itself ingested. Every product claim is backed by a failing CI test (claim-to-gate matrix: 99 GATED / 9 PARTIAL / 0 NO GATE). The latest GitHub release with notes and downloads: <https://github.com/beenuar/AiSOC/releases/latest>.+> **TL;DR for first-time visitors:** AiSOC is on `v8.1.0`, released 2026-09-23 — wave-2 features plus the gaps behind them. Every wave-2 backlog item was audited against the tree before any code was written, and the backlog was wrong in both directions: two items were already built, four had the capability present with the path that feeds it broken. One inverted what the feature appeared to do — a Slack or Teams approval authorized nobody. Every product claim is backed by a failing CI test (claim-to-gate matrix: 99 GATED / 9 PARTIAL / 0 NO GATE). The latest GitHub release with notes and downloads: <https://github.com/beenuar/AiSOC/releases/latest>.---## What's new-`VERSION` is `8.0.0`. The **v8.0.0** release (2026-09-22) is the **Close the loop** release. v8.0 had been reserved for the package-publish milestone; nothing can publish without registry credentials, so the milestone was re-scoped and distribution/packaging moved out. What v8.0 does instead is close the gap between what the codebase contains and what it actually runs.+`VERSION` is `8.1.0`. The **v8.1.0** release (2026-09-23) delivers the wave-2 backlog and the defects auditing it surfaced.++**v8.1.0 highlights (September 23, 2026)**+- **A ChatOps approval authorized nobody.** The Slack and Teams bots verified who clicked — Slack signs every interaction payload, Teams payloads carry an HMAC — recorded that person in an audit event, and then called the actions service with no approver. The permission-tier check and separation of duties were both skipped. A bot cannot supply permissions (it knows a Slack user id and has no idea what that person may do in AiSOC), so it now asserts identity only and the actions service maps it through operator configuration. **If you use ChatOps approvals you must populate `AISOC_CHATOPS_APPROVERS` or they will be refused** — that is the intended failure.+- **The signed email-approval fallback linked to a 404.** `approval_url()` pointed at a path no router served, so the documented answer to "Slack is unreachable" failed at the moment it was needed. The route exists, and the recipient is signed into the token, because a bare signed link is a bearer credential that approves as nobody.+- **The API service had no LLM input contract at all.** Seven endpoints POSTed untrusted input straight to a provider, including a submitted email body. The rules are now shared with `services/agents` rather than reimplemented, and the no-bypass gate — which walked the AST for `.ainvoke`/`.astream` and so could not see a raw-HTTP call — has a second half.+- **One `not` rule silently discarded a tenant's entire business-context rule set** at triage, suppressions included, because two evaluators disagreed about whether `not` takes a mapping or a list.+- **A rule id named one rule in the engine and a different one in the catalogue** for 45 network rules, so an analyst looking up an id off an alert read the wrong rule's description and playbook.+- **First-run fixes.** The Codespaces quickstart could never start Docker (capabilities are granted at container creation and cannot be self-granted from an image), and service images were published amd64-only, so Apple Silicon could not pull a single one.++Packaging moves to **v8.2**. The blocker is registry credentials, not code — `release.yml` already builds, packs and would upload all eight packages — and a release cannot schedule an account action by writing a version number.++The full inventory lives under `[8.1.0]` in [`CHANGELOG.md`](CHANGELOG.md).++---++## v8.0.0 (2026-09-22)++`VERSION` was `8.0.0`. The **v8.0.0** release (2026-09-22) is the **Close the loop** release. v8.0 had been reserved for the package-publish milestone; nothing can publish without registry credentials, so the milestone was re-scoped and distribution/packaging moved out. What v8.0 does instead is close the gap between what the codebase contains and what it actually runs.**v8.0.0 highlights (September 22, 2026)**- **The finding worth remembering, because it repeated a dozen times:** the mechanism existed, was unit-tested, and had no caller on the path that needed it. A passing test on an uncalled function is indistinguishable from a working feature until someone traces the call graph. `evidence_fingerprint` promised volatile fields were excluded while its only caller hashed the alert row id plus the whole raw event, so no two alerts ever matched and v7.7's repeat-alert suppression could only ever report zero. `PostActionVerifier` had no caller, and its isolation probe returned `bool(device_id)` — it would have certified an uncontained host. The console wrote per-tenant L0–L4 autonomy tiers to Postgres while the dispatcher read one global environment variable. `get_entity_neighbors` accepted a `tenant_id` and never passed it to the driver.diff --git a/ROADMAP.md b/ROADMAP.mdindex ffe397056..69e5b6054 100644--- a/ROADMAP.md+++ b/ROADMAP.md@@ -599,9 +599,16 @@ Disposition of every item that had been listed against v8.0:----## v8.1 — In progress+## v8.1 — Shipped (2026-09-23)-Wave-2 features plus release integrity. Tracked in+Wave-2 features plus release integrity. Every backlog item was audited against+the tree before any code was written, and the backlog was wrong in both+directions: two items were already built, and four had the capability present+with the path that feeds it broken. Two remain partial and say so — the two+attack-chain implementations still never exchange data, and ChatOps still has+no proactive card push or durable approval store.++Full inventory under `[8.1.0]` in [`CHANGELOG.md`](CHANGELOG.md). Tracked in[`docs/roadmap/v8-progress.md`](docs/roadmap/v8-progress.md) and, for thecommunity-facing view, issue[#362](https://github.com/beenuar/AiSOC/issues/362).diff --git a/VERSION b/VERSIONindex ae9a76b92..8104cabd3 100644--- a/VERSION+++ b/VERSION@@ -1 +1 @@-8.0.0+8.1.0diff --git a/apps/web/package.json b/apps/web/package.jsonindex e57bb0030..e05c1fb81 100644--- a/apps/web/package.json+++ b/apps/web/package.json@@ -1,6 +1,6 @@{"name": "@aisoc/web",- "version": "8.0.0",+ "version": "8.1.0","private": true,"description": "AiSOC SOC Console - Next.js 14 Frontend","license": "MIT",diff --git a/docs/roadmap/v8-progress.md b/docs/roadmap/v8-progress.mdindex d3429f9f4..325fbf4a4 100644--- a/docs/roadmap/v8-progress.md+++ b/docs/roadmap/v8-progress.md@@ -1,7 +1,7 @@# AiSOC v8 — progress tracker**Last updated:** 2026-09-23-**Current release:** `v8.0.0` (2026-09-22) · **In progress:** `v8.1`+**Current release:** `v8.1.0` (2026-09-23) · **Next:** `v8.2` (packaging)This tracker is the at-a-glance view of what has landed across the v8 line andwhat is still open. It is deliberately short and dated. When it disagrees with@@ -54,7 +54,7 @@ not code.----## v8.1 — in progress+## v8.1 — shipped 2026-09-23### Wave-2 features (issue [#362](https://github.com/beenuar/AiSOC/issues/362))@@ -92,7 +92,7 @@ optional one.----## Not in v8.1: packaging+## Not in v8.1 either: packaging`release.yml` already builds, packs and would upload all eight packages onevery tag. The repository's only secret is `FLY_API_TOKEN` — there is nodiff --git a/packages/aisoc-sandbox/tests/test_determinism.py b/packages/aisoc-sandbox/tests/test_determinism.pyindex 48701e3b9..6a61bc42b 100644--- a/packages/aisoc-sandbox/tests/test_determinism.py+++ b/packages/aisoc-sandbox/tests/test_determinism.py@@ -36,12 +36,27 @@ SCENARIOS = [# Wall-clock fields are expected to differ between runs. Everything else is# reasoning output and must not.-VOLATILE = {"latency_ms", "timestamp_utc", "total_latency_ms", "duration_ms"}+#+# This was an exact-name list of four fields, none of which the CLI emits: the+# actual field is `elapsed_ms`. So the test compared a timer and failed+# whenever two runs happened to straddle a millisecond boundary — reporting+# "Something in the reasoner depends on salted hashing again", which sends a+# reader to look for a `hash()` call that is not there.+#+# Matched by suffix rather than by name so a newly-added timing field cannot+# reintroduce the flake. Names are the thing that drifts; the `_ms` convention+# is not.+VOLATILE = {"timestamp_utc", "started_at", "finished_at"}+VOLATILE_SUFFIXES = ("_ms", "_seconds", "_at", "_duration")+++def _is_volatile(key: str) -> bool:+ return key in VOLATILE or key.endswith(VOLATILE_SUFFIXES)def _strip_volatile(value: Any) -> Any:if isinstance(value, dict):- return {k: _strip_volatile(v) for k, v in value.items() if k not in VOLATILE}+ return {k: _strip_volatile(v) for k, v in value.items() if not _is_volatile(k)}if isinstance(value, list):return [_strip_volatile(v) for v in value]return value@@ -67,7 +82,9 @@ def test_reasoning_is_identical_under_different_hash_seeds(scenario: str) -> Nonsecond = _run(scenario, "987654")assert json.dumps(first, sort_keys=True) == json.dumps(second, sort_keys=True), (f"{scenario} produced different reasoning under a different PYTHONHASHSEED. "- f"Something in the reasoner depends on salted hashing again."+ f"Something in the reasoner depends on salted hashing again — look for "+ f"hash(), set iteration order, or dict ordering derived from one. If the "+ f"only difference is a timer, add its field to VOLATILE instead.")