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): v20.2.1 (#1542)
Beenu Arora73ac0d295d92
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 27cd8cf54..0a6a7d8c9 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -7,6 +7,33 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0## [Unreleased]+## [20.2.1] - 2026-10-28++### Security++- **ReDoS in phishing header analysis**+ ([GHSA-9m2j-23jf-gg3p](https://github.com/beenuar/AiSOC/security/advisories/GHSA-9m2j-23jf-gg3p),+ medium, reported by [q7256](https://github.com/q7256); affects+ v19.0.0 – v20.2.0). `_ADDRESS_IN_NAME` in+ `services/api/app/services/phishing/headers.py` was an unanchored scan,+ so a long `From` display name with no address-like substring was+ re-consumed from every interior position — quadratic, measured at 33 s+ for 128 KB and reported at 71 s, and `POST /api/v1/phishing/reports`+ (authenticated, `cases:write` — the default analyst role) runs it+ synchronously on the event loop, stalling every other request on the+ process. Two independent bounds: a `(?<![\w.+-])` lookbehind makes+ interior starts fail in O(1) with the match set unchanged, and the+ display name is capped at 1,024 characters before analysis — which also+ keeps the impersonation finding's stored `evidence` out of the megabyte+ range on a crafted report. 128 KB now completes in 26 ms, linear. The+ regression test pins a 1-second budget with ≥30× margin both ways, was+ demonstrated failing on the pre-fix tree, and a reverted-fix negative+ control was demonstrated failing again. Every other `re.compile` in the+ phishing package was audited for the same class; the one suspicious+ sibling (`campaigns.py` `_OPAQUE`) was measured linear — its leading+ `\b` already prevents interior-start rescans — and deliberately left+ unchanged.+## [20.2.0] - 2026-10-28### Securitydiff --git a/README.md b/README.mdindex ed49214a1..fd095469c 100644--- a/README.md+++ b/README.md@@ -6,7 +6,7 @@**An open-source, self-hostable AI Security Operations Center.** It ingests your security telemetry, detects and correlates threats, investigates them with AI agents whose reasoning is fully auditable, and proposes responses a human approves.-[](https://opensource.org/licenses/MIT) [](CHANGELOG.md) [](https://github.com/beenuar/AiSOC/actions/workflows/ci.yml)+[](https://opensource.org/licenses/MIT) [](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) [](https://raw.githubusercontent.com/beenuar/AiSOC/main/apps/web/public/papers/aisoc-technical-guide.pdf)**[Technical Guide (PDF)](https://raw.githubusercontent.com/beenuar/AiSOC/main/apps/web/public/papers/aisoc-technical-guide.pdf)** · [Docs](https://beenuar.github.io/AiSOC/) · [Architecture](docs/architecture/README.md) · [What actually works](docs/audit/REPOSITORY_REALITY.md) · [Discussions](https://github.com/beenuar/AiSOC/discussions)diff --git a/RELEASES.md b/RELEASES.mdindex 85f047473..e545adb54 100644--- a/RELEASES.md+++ b/RELEASES.md@@ -2,7 +2,9 @@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 `v20.2.0`, released 2026-10-28. **The first-run-truth release: a fresh Windows install could not boot — and nothing could see it — and the install documentation described a different product; both are fixed, with gates so neither drifts back.** `install.ps1` initialised `.env` with a plain copy of the template, which provisions an environment no real deployment would boot (`ENVIRONMENT=production` with four empty secrets is exactly what `enforce_secure_defaults` refuses), and a Postgres already on 5432 or an Ollama on 11434 — the two conflicts most likely in this audience's environment — ended the install with advice to hand-edit `docker-compose.yml`, while `README.md` promised the opposite. Both paths now run the same generators `make up` runs (`scripts/ensure_env.py`, `scripts/resolve_port_conflicts.py`), a `windows-latest` CI job proves every generated secret lands non-empty, and the installer-parity gate reads the installer's action lines rather than its help text — three of its assertions had been passing on a script that merely *mentioned* the commands. The docs sweep corrected a quickstart advertising a machine 1.5 GB smaller than the measured requirement, an installation page saying "fourteen" CORE containers (sixteen — the number word survived every digit-based check, so the profile-count gate now reads words too), and four surfaces advertising Swagger at an address the production-class stack answers 404 on. The on-ramps consolidated to ones that can prove themselves: `render.yaml` is removed (a paid one-click path to a populated console that cannot answer whether the pipeline works — no ingest, no fusion, Kafka disabled), and the devcontainer attach banner and sandbox README stopped pointing at the demo compose file whose own header says not to evaluate with it. The Jira connector moved to `GET /rest/api/3/search/jql` after Jira Cloud removed the old endpoint (410 Gone on every poll); the nightly reproducible-builds aggregator stopped reading its own matrix's designed skip as lost coverage. The image `stable`-channel guard that had **failed open** across the v19 → v20 application major — a buildx read came back empty, and "empty" was read as "first release" — now reads the registry HTTP API with three states, where a tag that is present but unreadable **blocks** the channel; the channel itself was restored to v19.0.0 by the dispatchable repair workflow, the recorded owner decision. `qs` moves to **6.17.0** via the repo-wide override ([GHSA-6pq8-jhm2-2vcw](https://github.com/advisories/GHSA-6pq8-jhm2-2vcw), medium), and the two unpatchable `deepmerge` alerts (LOW; the latest published release is inside the vulnerable range) are dismissed on recorded evidence — their only dependents are dev/build tooling merging repo-authored configs. **11 defects were fixed in this release, 34 across it and the two before** (11 here, 11 in v20.1.0, 12 in v20.0.0 — counting distinct product or CI defects named in each release's Fixed and Security sections, dependency-version advisories excluded). No breaking changes, and **neither `stable` channel moves**. The chart publishes as **`7.11.0`**. Everything below describes v20.1.0 and earlier, which this release does not change.+> **TL;DR for first-time visitors:** AiSOC is on `v20.2.1`, released 2026-10-28. **A security release, no breaking changes.** It closes [GHSA-9m2j-23jf-gg3p](https://github.com/beenuar/AiSOC/security/advisories/GHSA-9m2j-23jf-gg3p), a medium-severity denial of service reported by [q7256](https://github.com/q7256): the phishing header analysis scanned a reported message's `From` display name with an unanchored regex, so a long display name containing no address-like substring was re-consumed from every interior position — quadratic, measured at 33 s for 128 KB on maintainer hardware (71 s in the reporter's measurement) — and `POST /api/v1/phishing/reports`, reachable by the default analyst role via `cases:write`, runs that scan synchronously on the event loop, so **one crafted phishing report stalled every other request on the API process**, including unauthenticated health checks. The fix applies two independent bounds: a lookbehind pins each match attempt to the start of a local-part run (interior starts now fail in O(1), match set unchanged), and the display name is analysed on its first kilobyte only (RFC 5322 caps a header line at 998 octets), which also keeps the finding's stored `evidence` out of the megabyte range. 128 KB now completes in 26 ms — ~1,270× faster, and linear. The regression test was demonstrated failing on the pre-fix tree and failing again under a reverted-fix negative control; every other regex in the phishing package was audited for the same class, and the one suspicious-looking sibling was *measured* linear and deliberately left unchanged. **If you run v19.0.0 through v20.2.0 and provision analyst accounts, upgrade.** 1 defect was fixed in this release, 23 across it and the two before (1 here, 11 in v20.2.0, 11 in v20.1.0 — counting distinct product or CI defects named in each release's Fixed and Security sections, dependency-version advisories excluded). The chart publishes as **`7.11.1`**, a patch because only `appVersion` moves. Neither `stable` channel moves. Everything below describes v20.2.0 and earlier, which this release does not change.+>+> **Previously:** AiSOC was on `v20.2.0`, released 2026-10-28. **The first-run-truth release: a fresh Windows install could not boot — and nothing could see it — and the install documentation described a different product; both are fixed, with gates so neither drifts back.** `install.ps1` initialised `.env` with a plain copy of the template, which provisions an environment no real deployment would boot (`ENVIRONMENT=production` with four empty secrets is exactly what `enforce_secure_defaults` refuses), and a Postgres already on 5432 or an Ollama on 11434 — the two conflicts most likely in this audience's environment — ended the install with advice to hand-edit `docker-compose.yml`, while `README.md` promised the opposite. Both paths now run the same generators `make up` runs (`scripts/ensure_env.py`, `scripts/resolve_port_conflicts.py`), a `windows-latest` CI job proves every generated secret lands non-empty, and the installer-parity gate reads the installer's action lines rather than its help text — three of its assertions had been passing on a script that merely *mentioned* the commands. The docs sweep corrected a quickstart advertising a machine 1.5 GB smaller than the measured requirement, an installation page saying "fourteen" CORE containers (sixteen — the number word survived every digit-based check, so the profile-count gate now reads words too), and four surfaces advertising Swagger at an address the production-class stack answers 404 on. The on-ramps consolidated to ones that can prove themselves: `render.yaml` is removed (a paid one-click path to a populated console that cannot answer whether the pipeline works — no ingest, no fusion, Kafka disabled), and the devcontainer attach banner and sandbox README stopped pointing at the demo compose file whose own header says not to evaluate with it. The Jira connector moved to `GET /rest/api/3/search/jql` after Jira Cloud removed the old endpoint (410 Gone on every poll); the nightly reproducible-builds aggregator stopped reading its own matrix's designed skip as lost coverage. The image `stable`-channel guard that had **failed open** across the v19 → v20 application major — a buildx read came back empty, and "empty" was read as "first release" — now reads the registry HTTP API with three states, where a tag that is present but unreadable **blocks** the channel; the channel itself was restored to v19.0.0 by the dispatchable repair workflow, the recorded owner decision. `qs` moves to **6.17.0** via the repo-wide override ([GHSA-6pq8-jhm2-2vcw](https://github.com/advisories/GHSA-6pq8-jhm2-2vcw), medium), and the two unpatchable `deepmerge` alerts (LOW; the latest published release is inside the vulnerable range) are dismissed on recorded evidence — their only dependents are dev/build tooling merging repo-authored configs. **11 defects were fixed in this release, 34 across it and the two before** (11 here, 11 in v20.1.0, 12 in v20.0.0 — counting distinct product or CI defects named in each release's Fixed and Security sections, dependency-version advisories excluded). No breaking changes, and **neither `stable` channel moves**. The chart publishes as **`7.11.0`**. Everything below describes v20.1.0 and earlier, which this release does not change.>> **Previously:** AiSOC was on `v20.1.0`, released 2026-10-22. **The harness-honesty release: the measurements and safety nets around the product were made as truthful as the product itself, and one real data-loss defect in every connector's poll loop is closed.** The data-loss fix is externally contributed ([meshreyy](https://github.com/meshreyy)): ingest reports per-event normalization failures as HTTP 200 with accepted/rejected counts, and the connector scheduler advanced its checkpoint on HTTP success alone — so a partially rejected batch moved the cursor past events that were never stored, permanently and silently. The cursor now advances only when every submitted event was accepted; a rejected page is re-fetched on the next overlapping poll and deduplicated by the sink. The live-agent eval was flattering itself two ways — per-incident `asyncio.run` left the model client on a dead event loop so most incidents silently ran heuristics, and failed incidents were dropped from the sample — it now runs one loop per run, scores failures as **0 with the reason recorded**, and the 0.40 groundedness floor was re-derived under the fixed harness (it holds, now 0.04 below the lowest honest run rather than 0.13 below a flattered one). The weekly chaos store-outage legs grade real behaviour instead of artifacts of their own probe contract, which is also why the API gained unauthenticated **`GET /api/v1/health` and `GET /api/v1/health/ready`** probes (database blocks at 503, cache degrades at 200) — the redis and qdrant legs pass end to end; the postgres leg stays red **by design**, naming the services that do not yet export the store keys the contract reads ([#1446](https://github.com/beenuar/AiSOC/issues/1446) stays open). The chart `stable` channel is now guarded by **both version lines** — v20.0.0 (chart 7.9.0, not a chart major) had carried chart `stable` across an application major while every image `stable` tag correctly held at v19 — and a dispatchable repair workflow exists and was used to roll the channel back to 7.8.0, where it stays until a deliberate promotion; **this release moves neither `stable` channel**. `fastapi` moves to 0.146.0 across all 32 declaration sites and 13 locks in one change, the Oct-22 Dependabot waves are folded, and `@vitest/mocker` 4.1.12 closes CVE-2026-112949/112950 in all three npm lockfiles. **11 defects were fixed in this release, 34 across it and the two before** (11 here, 12 in v20.0.0, 11 in v19.0.0 — counting distinct product or CI defects named in each release's Fixed and Security sections, dependency-version advisories excluded). The chart publishes as **`7.10.0`**. Everything below describes v20.0.0 and earlier, which this release does not change.>@@ -26,7 +28,27 @@ This file mirrors what used to live in the "What's new" section of [`README.md`]## What's new-`VERSION` is `20.2.0`. The **v20.2.0** release (2026-10-28) makes first+`VERSION` is `20.2.1`. The **v20.2.1** release (2026-10-28) is a+security patch closing+[GHSA-9m2j-23jf-gg3p](https://github.com/beenuar/AiSOC/security/advisories/GHSA-9m2j-23jf-gg3p)+(medium, reported by [q7256](https://github.com/q7256)): a quadratic+regex in the phishing header analysis let one crafted report — a long+`From` display name with no address-like substring, submitted through+`POST /api/v1/phishing/reports` by any holder of `cases:write`, which+the default analyst role has — stall the whole API process for tens of+seconds, because the scan runs synchronously on the event loop. The scan+is now linear (a lookbehind pins match attempts to run starts, with the+match set unchanged) and the display name is capped at 1 KB before+analysis, which also bounds what the impersonation finding stores as+evidence. 128 KB: 33 s → 26 ms. No breaking changes; neither `stable`+channel moves; the chart publishes as `7.11.1` (a patch — only+`appVersion` moves). If you run v19.0.0 – v20.2.0 and provision analyst+accounts, upgrade. See the `[20.2.1]` section of+[`CHANGELOG.md`](CHANGELOG.md) for the full inventory.++### Previously: v20.2.0++The **v20.2.0** release (2026-10-28) makes firstrun on Windows real — `install.ps1` generates `.env` secrets with thesame generator `make up` uses instead of copying a template the APIrefuses to boot on, resolves port conflicts instead of exiting withdiff --git a/VERSION b/VERSIONindex 86d4688d0..22eaaae7d 100644--- a/VERSION+++ b/VERSION@@ -1 +1 @@-20.2.0+20.2.1diff --git a/apps/docs/static/data/scoreboard.json b/apps/docs/static/data/scoreboard.jsonindex 01ea0715c..bd9ee09c7 100644--- a/apps/docs/static/data/scoreboard.json+++ b/apps/docs/static/data/scoreboard.json@@ -3,9 +3,9 @@"_readme": "Append-only public benchmark scoreboard. Each row is one eval run. Substrate rows (no LLM, deterministic) are marked substrate:true and MUST NOT be quoted as live agent performance. Wet-eval rows (substrate:false) are appended by the T5.5 weekly CI workflow once it lands. Newest row first.","rows": [{- "date": "2026-10-27",- "agent_version": "v20.2.0",- "commit_sha": "dd866be1",+ "date": "2026-10-28",+ "agent_version": "v20.2.1",+ "commit_sha": "a0a6ac3d","substrate": true,"eval_mode": "substrate-only","mitre_accuracy": 0.97,diff --git a/apps/docs/static/llms-full.txt b/apps/docs/static/llms-full.txtindex 110becfd7..7dfa83754 100644--- a/apps/docs/static/llms-full.txt+++ b/apps/docs/static/llms-full.txt@@ -3,7 +3,7 @@<!-- Generated by scripts/generate_llms_txt.py. Do not edit by hand; run the script and commit the result. -->> AiSOC is an open-source, self-hostable AI Security Operations Center, MIT-> licensed and maintained by the AiSOC community. Version 20.2.0.+> licensed and maintained by the AiSOC community. Version 20.2.1.> Repository: https://github.com/beenuar/AiSOC. Documentation: https://beenuar.github.io/AiSOC/.This file exists so an automated reader can answer questions about AiSOC@@ -127,7 +127,7 @@ plans/ Historical prototype subtree — reference only, never editedEvery number in this file is generated from the artefact that owns it. If youneed to state one, re-derive it rather than quoting a figure from prose —-including from this file, which is a snapshot of version 20.2.0.+including from this file, which is a snapshot of version 20.2.1.```python3 scripts/generate_corpus_stats.py --check # executable + on-disk rulesdiff --git a/apps/docs/static/llms.txt b/apps/docs/static/llms.txtindex 34cd4c7ba..c897160d3 100644--- a/apps/docs/static/llms.txt+++ b/apps/docs/static/llms.txt@@ -6,7 +6,7 @@> security telemetry from 87 first-party connectors, normalises it to OCSF,> correlates it into alerts, and runs LLM agents that triage and investigate them.> Every agent decision is written to an investigation ledger, so the reasoning-> behind a verdict can be replayed step by step. MIT licensed. Current version 20.2.0.+> behind a verdict can be replayed step by step. MIT licensed. Current version 20.2.1.AiSOC is built and maintained by the AiSOC community. It is a complete SOCplatform rather than a single tool: ingestion, detection, correlation, agentdiff --git a/apps/web/package.json b/apps/web/package.jsonindex 4c1ce576e..7502a1814 100644--- a/apps/web/package.json+++ b/apps/web/package.json@@ -1,6 +1,6 @@{"name": "@aisoc/web",- "version": "20.2.0",+ "version": "20.2.1","private": true,"description": "AiSOC SOC Console - Next.js 14 Frontend","license": "MIT",diff --git a/infra/helm/aisoc/Chart.yaml b/infra/helm/aisoc/Chart.yamlindex ab977f188..4cbfd6763 100644--- a/infra/helm/aisoc/Chart.yaml+++ b/infra/helm/aisoc/Chart.yaml@@ -25,13 +25,13 @@ type: application# the chart refuses grants it used to accept (see the v14.0.0 BREAKING# section), and the chart's major is the only coordinate a consumer pulling# `oci://ghcr.io/beenuar/charts/aisoc` sees before they install it.-version: 7.11.0+version: 7.11.1# Used as the image tag wherever a `tag:` in values.yaml is left empty, so# it has to be a tag the registry actually holds — it read "5.2.0", which# exists for no image, so every pod of a default install would have gone to# ImagePullBackOff. scripts/check_published_images.py resolves it the same# way the templates do and fails when it names nothing.-appVersion: "v20.2.0"+appVersion: "v20.2.1"keywords:- security