diff --git a/compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md b/compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md index 960dfb918..417dd886c 100644 --- a/compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md +++ b/compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md @@ -101,7 +101,7 @@ Threshold IDs follow the pattern `THR--`: | **THR-SEC-002** | OPA Circuit Breaker — Max Policy Evaluation Failures | 5 | count | IR-4, SC-7, SI-7 | OPA config in `deployment/opa_config.yaml`; `CircuitBreaker` in [`src/gateway/core/policy.py`](../../src/gateway/core/policy.py) (`failure_threshold=5`, `recovery_timeout=30.0`) | `lula-validation-cm6.yaml` | Structured gateway logs (`OPA circuit breaker: ... -> OPEN`); Prometheus metric planned | — | After 5 consecutive upstream failures (transport errors, HTTP 5xx, or HTTP 408/429), the circuit breaker opens for 30s and fast-fails to `"DENY"` before probing with a single `HALF_OPEN` request. Every failed call during the pre-open window also returns `"DENY"` (fail-closed); HTTP 4xx and local pool saturation (`httpx.PoolTimeout`) deny that request without tripping the breaker. | | **THR-AUD-001** | Audit Log Retention Period | 90 | days | AU-11, AU-12 | GCS lifecycle policy in `infra/`; Langfuse trace retention configured in Langfuse admin | `lula-validation-au12.yaml` (AU-12 audit record generation) | `cage.audit.log_age_days` (gauge — alert if logs deleted early) | POAM-003 (AU-12 synthetic traces) | 90 days is the minimum for HIGH-impact systems per AU-11. Organization retains for 1 year for financial regulatory compliance. | | **THR-AUD-002** | Lula Audit Run Cadence | 6 | hours | CA-7, AU-12 | Lula CronJob (`0 */6 * * *` standalone Terraform CronJob tracked under `POAM-2026-195`, retiring `385e502e:src/compliance_bridge/lula_scheduler.py` and `de5534ac:deployment/k8s/lula-cron.yaml`) | Self-validating (Lula run produces OSCAL assessment result) | `cage.lula.last_run_age_seconds` (gauge) | `POAM-2026-195` | 6-hour cadence provides near-real-time continuous monitoring per CA-7. | -| **THR-AUD-003** | SBOM Generation Cadence | 24 | hours | CM-8 | CI daily schedule in `.github/workflows/security-scan.yml` and `.github/workflows/sbom.yml` (`0 2 * * *`; unmanaged `de5534ac:deployment/k8s/sbom-cronjob.yaml` retired) | `lula-validation-cm6.yaml` (CM-8 component inventory) | `cage.sbom.last_generated_age_hours` (gauge) | POAM-006 | Daily CI SBOM generation and Artifact Registry image digests provide CM-8 component inventory. | +| **THR-AUD-003** | SBOM Generation Cadence | 24 (Python/FS) / per-push (images) | hours / event | CM-8 | Daily `0 2 * * *` + PR/push in [`.github/workflows/security-scan.yml`](../../.github/workflows/security-scan.yml) (Python `uv.lock` deps + repo filesystem); PR/push-only in [`.github/workflows/sbom.yml`](../../.github/workflows/sbom.yml) (4 container images; no `schedule` cron; unmanaged `de5534ac:deployment/k8s/sbom-cronjob.yaml` retired in `#686`, zero SBOM/CVE CronJobs in `infra/`) | — (`lula-validation-cm6.yaml` checks config settings only; no active CM-8 Lula spec) | — (CI workflow artifacts only; no runtime `cage.sbom.*` gauge) | POAM-006 (Open), POAM-2026-010 (Open), POAM-2026-221 | Daily CI Python/filesystem SBOM + Grype scan (`security-scan.yml`, restored in `#768` `b92a1719`) and build-time container SBOMs (`sbom.yml`); periodic in-cluster container scanning remains Open (`POAM-2026-010`). | | **THR-OBS-001** | OpenTelemetry Trace Sampling Rate | 0.01 (1%) | probability | AU-2, AU-12, SA-8 | OTel SDK config in `src/governed_financial_advisor/infrastructure/telemetry/__init__.py`; `OTEL_TRACES_SAMPLER_ARG=0.01` env var | `lula-validation-au12.yaml` | `cage.otel.trace_sample_rate` (gauge) | — | **KNOWN RISK ACCEPTANCE** — see Section 5. 1% sampling is insufficient for full AU-12 audit evidence. Compensating control: all governance decisions (policy evaluations, confidence checks, trade blocks) are logged at 100% regardless of sampling rate. | | **THR-OBS-002** | Langfuse Evidence Age Maximum (ISO 42001 A.5.2) | 172,800 | seconds (48 hr) | ISO 42001 A.5.2, AU-12, CA-7 | Freshness SLA monitor in `src/compliance_bridge/sla_monitor.py` (`evidence_age_seconds < 172800`) | `lula-validation-cm6.yaml` | `cage.langfuse.evidence_age_seconds` (gauge) | — | 48-hour evidence freshness window ensures Lula can validate continuous monitoring evidence. | | **THR-OBS-003** | ISO 42001 A.5.2 Safety Rate Minimum | 0.99 (99%) | rate [0,1] | ISO 42001 A.5.2, RA-3 | Lula Rego in `compliance/lula/lula-validation-a52.yaml` → `safety_rate >= 0.99` | Lula A.5.2 validation manifest (`lula-validation-a52.yaml`) | `cage.iso42001.a52_safety_rate` (gauge) | — | 99% of all inference requests must pass safety checks. Derived from ISO 42001 A.5.2 monitoring requirements. | @@ -150,10 +150,10 @@ This section provides a reverse lookup: for each NIST SP 800-53 control family, | **AU-12** | Audit Record Generation | Lula cadence, OTel sampling, evidence age freshness | THR-AUD-002, THR-OBS-001, THR-OBS-002 | ⚠️ PARTIAL — POAM-003 (synthetic traces); AU-12 validated by Lula | | **CM-6** | Configuration Settings | All thresholds in governance_thresholds.json are CM-6 configuration settings | ALL THR-\* | ✅ Validated by `schemas/thresholds.py` Pydantic enforcement + Lula | | **CM-7** | Least Functionality | OPA circuit breaker (fail-closed) prevents unauthorized gateway functionality on policy engine failure | THR-SEC-002 | ✅ Implemented — fail-closed circuit breaker | -| **CM-8** | System Component Inventory | SBOM generation cadence | THR-AUD-003 | 🔄 IN PROGRESS — POAM-006; remediated by sbom-cronjob.yaml | +| **CM-8** | System Component Inventory | SBOM generation cadence | THR-AUD-003 | ⚠️ PARTIAL — POAM-006 (Open); CI workflows (`sbom.yml` on PR/push, `security-scan.yml` daily + PR/push; `de5534ac:deployment/k8s/sbom-cronjob.yaml` was unmanaged and retired in `#686`) | | **IR-4** | Incident Handling | Drawdown thresholds trigger automated incident response (trade halt) | THR-FIN-002, THR-FIN-004 | ✅ Implemented — STPA validator auto-halt | | **RA-3** | Risk Assessment | All financial guardrail thresholds derive from RA-3 risk assessment | THR-FIN-001 through THR-FIN-006, THR-CBF-001 | ✅ Threshold values documented with risk rationale | -| **RA-5** | Vulnerability Monitoring | SBOM generation (CM-8/RA-5 overlap) enables CVE cross-referencing | THR-AUD-003 | 🔄 IN PROGRESS — POAM-010; Grype integration added | +| **RA-5** | Vulnerability Monitoring | SBOM generation (CM-8/RA-5 overlap) enables CVE cross-referencing | THR-AUD-003 | ⚠️ PARTIAL — CI `pip-audit` + Trivy/Syft/Grype (`POAM-010`, Python SBOM leg restored `2026-11-15` under `POAM-2026-221`); in-cluster periodic scanning Open under `POAM-2026-010` | | **SA-8** | Security Engineering | Confidence threshold, latency SLAs, CBF parameter reflect security engineering principles | THR-CONF-001, THR-LAT-001, THR-CBF-001 | ✅ Documented with STPA and formal methods derivation | | **SC-7** | Boundary Protection | OPA circuit breaker enforces boundary between external input and trade execution | THR-SEC-002 | ✅ Implemented | | **SC-8** | Transmission Confidentiality | Latency SLAs indirectly enforce SC-8 (slow responses may indicate TLS inspection); latency anomaly detection | THR-LAT-001 | ⚠️ PARTIAL — POAM-011 (no TLS assertion test) | @@ -257,22 +257,22 @@ This section documents thresholds that represent **known risk acceptances** — --- -### 5.3 RA-003: SBOM Scanning Latency — Daily vs. Real-Time (THR-AUD-003) +### 5.3 RA-003: SBOM & Vulnerability Scanning Cadence Gap — No In-Cluster CronJob / Build-Only Container SBOMs (THR-AUD-003) | Field | Value | | ------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | **Risk Acceptance ID** | RA-003 | | **Threshold** | THR-AUD-003 — SBOM Generation Cadence | -| **Current Value** | 24 hours (daily at 02:00 UTC) | -| **Ideal Value** | Per-build SBOM generation (on every CI push) | -| **Risk Accepted** | A new CVE disclosed after a container image is built but before the next daily SBOM scan (up to 23h 59m) would not be detected until the next scheduled scan. In a zero-day scenario, CAGE could run vulnerable code for up to 24 hours without automated detection. | -| **Why Not Per-Build?** | Full container image SBOM scanning (Syft + Grype) takes 3–8 minutes per image. The K8s CronJob scans all 4 CAGE images; doing this on every git push would add 15–30 minutes to CI and create GCS cost concerns for artifact storage. Per-build Python SBOM is implemented in CI (mitigates Python-specific risk). | +| **Current Value** | Python `uv.lock` & repo filesystem: daily (`0 2 * * *`) + PR/push in [`.github/workflows/security-scan.yml`](../../.github/workflows/security-scan.yml). Shipped container images (4 images): PR/push-only in [`.github/workflows/sbom.yml`](../../.github/workflows/sbom.yml) (no `schedule` cron; zero in-cluster SBOM/CVE `CronJob` in Terraform) | +| **Ideal Value** | Per-build SBOM generation plus scheduled daily re-scan of deployed container image digests | +| **Risk Accepted** | A CVE newly disclosed in an OS/container package after the most recent `main` image build is not automatically re-scanned against deployed container images until the next `sbom.yml` push/PR run or manual dispatch, because no periodic container-image `schedule` or in-cluster Kubernetes scanning `CronJob` is deployed (`POAM-2026-010`, Open). | +| **Correction (`2026-11-15`)** | Earlier versions of `RA-003` asserted (a) an active Kubernetes `CronJob` scanning all 4 CAGE images daily at `02:00 UTC` ("24h max" container exposure window) and (b) per-push Python dependency SBOM coverage via `security-scan.yml::sbom-generation`. Both assertions were inaccurate: **(a)** `de5534ac:deployment/k8s/sbom-cronjob.yaml` was an unmanaged pre-Terraform manifest retired in `#686` and Terraform (`infra/`) deploys no SBOM or vulnerability-scanning `CronJob` (`POAM-2026-010`, Open); **(b)** from `1902c92c` (`2026-06-01`) through `9b2b745d` (`2026-11-15`), `security-scan.yml::sbom-generation` ran `pip install cyclonedx-bom` without `uv sync` and emitted a 0-dependency Python SBOM (`POAM-2026-221`, remediated in `#768` `b92a1719` with `uv sync` + `expected_uv_lock_component_floor()`). | | **Who Accepted** | Authorizing Official (AO) — [SIGNATURE PENDING] | -| **Acceptance Date** | Pending AO signature (target: 2026-04-30) | -| **Compensating Controls** | (1) Python dependency SBOM IS generated on every CI push via `sbom-generation` GitHub Actions job. (2) pip-audit runs on every PR. (3) Trivy filesystem scan runs on every PR. (4) GitHub Dependabot alerts provide near-real-time CVE notification independent of SBOM pipeline. | -| **Residual Risk** | Low — container image CVE exposure window is 24h max; Python dependency exposure is near-zero due to per-build scanning. | -| **Review Date** | 2026-09-30 | -| **POAM Ref** | POAM-006 (being remediated), POAM-010 | +| **Acceptance Date** | Pending AO signature | +| **Compensating Controls** | (1) `security-scan.yml::python-dependency-audit` (`uv sync --frozen --all-groups --all-extras` + `pip-audit`) and `security-scan.yml::sbom-generation` (restored in `#768` `b92a1719`, `POAM-2026-221`) run on every PR, `main` push, and daily at `02:00 UTC`. (2) `sbom.yml` builds and Trivy-scans all 4 Wolfi container images on PRs/pushes touching `src/`, `Dockerfile*`, `pyproject.toml`, or `uv.lock`. (3) Trivy filesystem scan runs on every PR and daily. (4) GitHub Dependabot alerts provide CVE notifications on `uv.lock`. | +| **Residual Risk** | Moderate — Python dependency exposure is bounded to ≤24h (`security-scan.yml` daily schedule), while deployed container image digests lack scheduled in-cluster or daily registry re-scanning until `POAM-2026-010` closes. | +| **Review Date** | 2026-12-31 | +| **POAM Ref** | `POAM-006` (Open), `POAM-010` (Closed on CI `pip-audit` + Trivy; Python SBOM leg qualified under `POAM-2026-221`), `POAM-2026-010` (Open — in-cluster CronJob), `POAM-2026-221` (Closed at `b92a1719`) | --- diff --git a/docs/POAM.md b/docs/POAM.md index 55679edf1..8ca5747f5 100644 --- a/docs/POAM.md +++ b/docs/POAM.md @@ -68,7 +68,7 @@ The following findings are tracked as open items with target remediation dates. | ID | Control | Description | Severity | Target Date | |----|---------|-------------|----------|-------------| -| POAM-2026-010 | RA-5 | In-cluster vulnerability scanning CronJob not yet deployed (deploy-time operational hygiene; build-time image scanning runs in [`sbom.yml`](../.github/workflows/sbom.yml) and [`security-scan.yml`](../.github/workflows/security-scan.yml)) | Low | 2026-09-30 | +| POAM-2026-010 | RA-5 / CM-8 | In-cluster vulnerability/SBOM scanning `CronJob` not yet deployed in Terraform (`de5534ac:deployment/k8s/sbom-cronjob.yaml` was an unmanaged shadow manifest retired in `#686`, and `infra/` deploys only `CronJob/lula-audit`). Build-time container image scanning runs on PR/push in [`sbom.yml`](../.github/workflows/sbom.yml) (no `schedule` cron), while daily `0 2 * * *` scanning in [`security-scan.yml`](../.github/workflows/security-scan.yml) covers locked Python dependencies and the repository filesystem only. **Update (`2026-11-15`):** corrected [`THRESHOLD_TRACEABILITY_MATRIX.md`](../compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md) (`THR-AUD-003`, `CM-8`, `RA-5`, and `§5.3 RA-003`) to withdraw its false assertions of an active daily K8s SBOM `CronJob`, a `24h` container-image CVE exposure window, and a non-existent `cage.sbom.last_generated_age_hours` gauge | Moderate | 2026-12-31 | | POAM-2026-011 | SC-8 | **Reopened (`2026-11-05` — Dead-code closure, `F-104`, `POAM-2026-119`).** Closed `2026-08-27` on [`create_hardened_client_context()`](../src/gateway/infrastructure/tls_context.py) and [`tests/test_tls_enforcement.py`](../tests/test_tls_enforcement.py), citing commit `c8d66189a94cb782f4e0f61e14e05e341d27a8f6` (which resolves nowhere in `git log --all`, `POAM-2026-119`). Zero production call sites in `src/` invoke `create_hardened_client_context()`: outbound HTTP clients (`httpx`, `urllib3`, `aiohttp`) construct default SSL contexts while [`redis_client.py`](../src/gateway/infrastructure/redis_client.py) enforces TLS 1.2+ independently on Redis connections only. Closes on wiring `create_hardened_client_context()` into production outbound clients or restating the OSCAL `SC-8` claim (`PR-7c`, detail section) | Moderate | 2026-12-31 | | POAM-2026-012 | SC-12 / IA-5 | **Reopened (`2026-11-05` — Prose-only closure, `PR-9` Check 1).** Closed `2026-08-22` citing only [`docs/operations/KEY_ROTATION.md`](operations/KEY_ROTATION.md) with zero runnable test, zero automated asymmetric Cloud KMS signing-key rotation (`gateway-seal`, `reconciler-snapshot`, `compliance-evidence`, `benchmark-signing` in [`kms_signing.tf`](../infra/targets/gcp-gke/kms_signing.tf)), and stub-only [`lula-validation-ia5.yaml`](../compliance/lula/lula-validation-ia5.yaml). Closes on runnable key-lifecycle / dual-`kid` rotation verification in pytest/Terraform tests and an active `IA-5` validation (`POAM-2026-197`) | Moderate | 2026-12-31 | | POAM-2026-025 | NIST AI 600-1 §2.6 | CBRN / harmful content Lula validation (`lula-validation-ai600-cbrn.yaml`) is a stub pending AO pre-approval for NeMo CBRN rail deployment | High | 2026-12-31 | @@ -123,7 +123,6 @@ The following findings are tracked as open items with target remediation dates. | POAM-2026-217 | AU-5 / AU-10 / AU-12 / AC-3 / ISO 42001 A.5.2, A.9.2 | **Silent evidence-loss and outage-inversion paths on stopped or broken `EvidenceStreamSink` across single-use seal consumption (`GOVERNANCE_SEAL_CONSUMED`) and permitted inference completions (`A.5.2`, `A.9.2`), masked by dual `app.state.evidence_sink` precedence and undetected by ingress request metrics.** **(1) Single-use seal burned without DB 1 consumption proof (`POAM-2026-156` audit half):** [`verify_and_consume_seal()`](../src/gateway/governance/routing_seal.py) gated its `GOVERNANCE_SEAL_CONSUMED` append and `is_evidence_chain_blocking()` check behind `if has_redis or getattr(sink, "is_running", False):`, so on a stopped sink (`_redis is None`, `is_running is False`) DB 0 burned the one-time seal nonce while DB 1 silently skipped `GOVERNANCE_SEAL_CONSUMED` without raising. **(2) Outage inversion on `POST /v1/chat/completions`:** while [`_emit_proxy_evidence()`](../src/gateway/server/inference_proxy.py) failed closed with HTTP `503` on refusals, [`_emit_proxy_pass_evidence()`](../src/gateway/server/inference_proxy.py) ignored `is_evidence_chain_blocking()`, returned silently when `not sink._running`, ignored `None` returns from `sink.ingest()`, and swallowed `EvidenceChainCorruptError` — blocking refusals while serving permitted completions and Stage 4 PII-redacted responses (`200 OK`) with zero `A.5.2`/`A.9.2` evidence. **(3) Undocumented dual sink precedence:** only `inference_proxy.py` read `request.app.state.evidence_sink` ahead of [`get_evidence_sink()`](../src/gateway/governance/evidence/stream.py), allowing a stopped sink on `app.state` to shadow the singleton. **(4) Blind spot in metrics:** `cage_gateway_gated_requests_total` incremented independently of Redis appends while `compliance-bridge` simply observed fewer traces and treated the cluster as idle. **Remediated in code (`2026-11-14`):** `verify_and_consume_seal()` and `_emit_proxy_pass_evidence()` route unconditionally through [`commit_decision_evidence()`](../src/gateway/governance/evidence/stream.py) under `is_evidence_chain_blocking()` (rolling back token quota and returning HTTP `503`), `app.state.evidence_sink` / `inference_app.state.evidence_sink` are retired, and [`google_monitoring_alert_policy.gateway_evidence_append_reconciliation`](../infra/targets/gcp-gke/metrics_alerting.tf) reconciles `cage_gateway_gated_requests_total{outcome="permitted"}` and `cage_gateway_evidenced_requests_permitted_total` against `cage_gateway_evidence_records_appended_total`, guarded by [`test_fail_closed_evidence_paths.py`](../tests/unit/test_fail_closed_evidence_paths.py); remains Open pending mainline squash-merge SHA (detail section) | High | 2026-11-30 | | POAM-2026-219 | CA-2 / CA-7 / SA-11 | **Lula manifests rated or cited beyond what their Rego checks.** `#664` (`372daad2`, 2026-11-06), a Lula v0.16 schema migration of all 33 manifests, removed assertions from `ftra`, `eu-fria`, `tqp007`, `flowsignal`, `iso001-token-quota` and `sc8` without changing their [`compliance/lula/README.md`](../compliance/lula/README.md) rows, their headers or the closures citing them, and its message describes a schema migration and no removals. `eu-fria` backed closed POAM-2026-099 while checking neither the `fria` tier nor the `EU_ECB` region. Remediated 2026-11-14: four manifests re-rated Stub, the `iso001-token-quota` policy-content check restored, the `sc8` header corrected, the citations in POAM-2026-030-B, -074, -080, -099, -167 and -195 withdrawn, [`test_lula_description_parity.py`](../tests/unit/test_lula_description_parity.py) holding every Active row to its Rego, and [`check_k8s_terraform_parity.py`](../scripts/check_k8s_terraform_parity.py) failing an Active manifest that selects an object `infra/` does not create. Scanning every ledger and the evidence-claim census found older citations of manifests already rated Stub, listed in the detail section for a later fix: closed POAM-2026-079 cites `ac2`, and seven US SSP statements cite `ac2`, `aarm-vectors` or an AI 600-1 placeholder. Remains Open pending the mainline squash-merge SHA, a Gate W run of the restored check, and those fixes (detail section) | Moderate | 2026-11-30 | | POAM-2026-220 | SC-7 / AU-9 / CA-7 / GDPR Art. 44 / MAS TRM §4.2 | **Latent `CAGE_ENV=prod` `compliance-bridge` startup failure in `resolve_cold_store_bucket()` masked by `staging` geographic-residency bypass.** [`resolve_cold_store_bucket()`](../src/gateway/governance/evidence/residency.py) skipped geographic validation whenever `CAGE_ENV` was in `("development", "dev", "test", "ci", "staging", "uat", "preprod")`, so Gate W never exercised runtime cold-store residency (`residency_result_breakdown` in [`gate_w_certification.json`](../compliance/gate_w_certification.json)). Under `CAGE_ENV=prod` (`US_FED`, `EVIDENCE_COLD_STORE=gcs`, `EVIDENCE_COLD_STORE_BUCKET=-evidence-worm-prod`, and no `GOOGLE_CLOUD_LOCATION` on `module.compliance_bridge`), `resolve_cold_store_bucket(backend_id="gcs")` rejected the unprefixed bucket with `ResidencyViolationError` and crashed `EvidenceCustodian.from_env()` at bridge startup, while the identical inputs passed under `CAGE_ENV=staging`; [`get_cold_store()`](../src/gateway/governance/evidence/factory.py) also bypassed residency whenever a caller passed `bucket=`. **Remediated in code (`2026-11-15`):** `resolve_cold_store_bucket()` enforces residency across all `is_enforcing()` postures (`staging` included) against `EVIDENCE_COLD_STORE_LOCATION` — injected into [`module.compliance_bridge`](../infra/modules/compliance_bridge/main.tf) from [`module.worm_bucket.bucket_location`](../infra/modules/worm_bucket/outputs.tf) — retiring bucket-name prefixes, `GOOGLE_CLOUD_LOCATION`/`AWS_REGION` fallbacks, and `LOCAL`, and [`get_cold_store()`](../src/gateway/governance/evidence/factory.py) passes `bucket_override=bucket`, guarded by [`test_evidence_residency.py`](../tests/test_evidence_residency.py) and [`operator_identity.tftest.hcl`](../infra/modules/compliance_bridge/tests/operator_identity.tftest.hcl). Closes on the mainline squash-merge SHA plus a Gate W run exercising `compliance-bridge` startup and WORM custody with residency active in `staging` (detail section) | High | 2026-11-30 | -| POAM-2026-221 | CA-2 / CA-7 / CM-8 / RA-5 / SA-11 | **Vacuous CI gates caused by narrow `pip install` runner environments and `raise_on_error=False` / `ContentCheck.UNAVAILABLE` silent fallback (`stpa-freshness-check`, `sbom-generate`, `sbom-generation`), plus committed 8-package scratch SBOM.** **(1) `ci.yml::stpa-freshness-check` (`49c0d09a` `#694` `2026-11-08` through `9b2b745d` `#767` `2026-11-15`, 70 merged PRs):** `#407` (`b4bb9ef7`, `2026-10-06`) provisioned `stpa-freshness-check` with `pip install pyyaml pydantic ruff` instead of `uv sync`; `#659` (`d8f2acd3`, `2026-11-05`) added the 3 `US_FED` OSCAL artifacts behind `render_default_oscal_artifacts(..., raise_on_error=False)` and `except Exception: pass` -> `ContentCheck.UNAVAILABLE` commit-order fallback; and `#694` (`49c0d09a`, `2026-11-08`) added `_live_registered_server_inputs()`, importing `opentelemetry` transitively and causing `stpa-freshness-check` to catch `ModuleNotFoundError: No module named 'opentelemetry'` and silently skip all 3 `US_FED` OSCAL content checks across 70 mainline merges while reporting green (`test_stpa_freshness_gate.py` in `pytest-logic` continued running under `uv sync` and held the committed files fresh). **(2) Workflow-wide `pip install` sweep & release/repo SBOM audit (`2026-11-15`):** `ci.yml::sbom-generate` (`8c6e9361` `#16`, `2026-06-15`–`2026-11-15`, 5 months) and `security-scan.yml::sbom-generation` (`1902c92c`, `2026-06-01`–`2026-11-15`, 5.5 months, cited by closed `POAM-010` in [`POAM_US_FED.md`](compliance/us_fed/POAM_US_FED.md)) installed only `cyclonedx-py` / `cyclonedx-bom` into unsynced runner environments, emitting ephemeral workflow SBOMs with zero CAGE dependencies (never attached to any of the 9 GitHub Releases, GHCR packages, attestations, S3/GCS, or partner docs; Gate W `8956f505` / `fa772122` also cites zero of the 3 vacuous jobs), while `9b2b745d:compliance/sbom/python-deps-2026-06-08.cdx.json` committed an 8-package scratch gRPC SBOM in the public repo. **Remediated in `#768` (`2026-11-15`):** `render_default_oscal_artifacts()` defaults `raise_on_error=True`, [`check_stpa_freshness.py`](../scripts/check_stpa_freshness.py) deletes the commit-order fallback, [`generate_sbom.py`](../scripts/generate_sbom.py) and `ci.yml::sbom-generate` enforce `expected_uv_lock_component_floor()` (`>= 90%` of platform-applicable `uv.lock` packages), the committed 8-package SBOM is deleted and withdrawn, all six affected workflow jobs run `uv sync`, and [`test_stpa_freshness_gate.py`](../tests/test_stpa_freshness_gate.py) + [`test_ci_workflow_dependency_sync.py`](../tests/unit/test_ci_workflow_dependency_sync.py) guard against recurrence; remains Open pending mainline squash-merge SHA (detail section) | High | 2026-11-30 | ### EU ECB Region (EU_ECB) @@ -314,6 +313,7 @@ The following findings have been remediated and verified via Lula validation and | POAM-2026-216 | SA-11 / CA-2 / CA-7 | ✅ **CLOSED — Synthetic `DEFAULT_JOB_LOGS` stub replaced with canonical `cage_lula_audit_receipt_v1` verification and independent bijective active-manifest stem parity (`set(satisfied_controls) == set(active_stems)`) across all 26 active Lula specs in Gate W Phase 3.** Merged in `#748` (`0ec2a7f7`); [`run_lula_audit.sh`](../infra/modules/lula_cron/run_lula_audit.sh) enforces closed-world `lula.assessment_remark` status tokens on `a52`/`a53`/`a92` and emits `cage_lula_audit_receipt_v1`, while [`verify_lula_gate_w_job_logs()`](../scripts/verify_lula_assessment_results.py) (`--lula-manifests-dir compliance/lula` in [`scripts/staging_lifecycle.sh`](../scripts/staging_lifecycle.sh)) enforces `startup_grace_count == 0` and exact stem-set equality against [`discover_active_lula_manifests()`](../scripts/verify_lula_assessment_results.py). Guarded by [`tests/unit/test_oscal_idle_trace_remarks.py`](../tests/unit/test_oscal_idle_trace_remarks.py) and [`tests/infrastructure/test_staging_lifecycle.py`](../tests/infrastructure/test_staging_lifecycle.py). | 2026-11-14 | | POAM-2026-165 | AC-6 / CM-5 / SI-7 | ✅ **CLOSED — Two-identity GitHub Actions WIF partition (`sa-cage-deployer` restricted to `attribute.ref/refs/heads/main` with `roles/editor`/`run.admin`/`secretmanager.admin` removed; `cage-partner-harness` isolated for partner wire runs), dedicated `cage-gke-nodes-staging` node GSA (`POAM-2026-187`), and `-compute@developer.gserviceaccount.com` Token Creator + project-level privilege revocation verified live on `` (`2026-11-14`).** Merged in `#661` (`c58b0913`, [`infra/targets/gcp-gke/wif.tf`](../infra/targets/gcp-gke/wif.tf)) and `#658` (`38e12586`, [`infra/modules/gcp_gke_cluster/main.tf`](../infra/modules/gcp_gke_cluster/main.tf)), exercised in clean-room Gate W `#738` (`84f39e21`, certifying `8956f505` in [`compliance/gate_w_certification.json`](../compliance/gate_w_certification.json): `227` image pulls / `115` cold pulls / `0` container restarts on GKE 1.35 under `roles/container.defaultNodeServiceAccount` + `roles/artifactregistry.reader` only), verified over the wire (`run 41944192524` `PERMISSION_DENIED` on non-`main` `sa-cage-deployer` impersonation; `run 41943975961` `7 passed` under `cage-partner-harness`) and live on `` (`2026-11-14`: `cage-staging` `1.36.4-gke.2377000` runs all four node pools under `cage-gke-nodes-staging@.iam.gserviceaccount.com`; `-compute@developer.gserviceaccount.com` absent from `sa-cage-deployer` and `cage-gateway` IAM policies and stripped to `roles/cloudtrace.agent` + `roles/logging.logWriter` only; [`scripts/bootstrap_gcp_project.sh`](../scripts/bootstrap_gcp_project.sh) strips default compute SA grants on fresh bootstrap), and guarded by [`tests/infrastructure/test_wif_deployer_isolation.py`](../tests/infrastructure/test_wif_deployer_isolation.py), [`tests/infrastructure/test_gke_node_pool_service_account.py`](../tests/infrastructure/test_gke_node_pool_service_account.py), [`infra/targets/gcp-gke/tests/w6.tftest.hcl`](../infra/targets/gcp-gke/tests/w6.tftest.hcl), and [`tests/infrastructure/test_gcp_project_bootstrap.py`](../tests/infrastructure/test_gcp_project_bootstrap.py). | 2026-11-14 | | POAM-2026-187 | AC-6 / AU-9 / AU-10 / CM-6 | ✅ **CLOSED — All six GKE `node_config` blocks bind dedicated `cage-gke-nodes-${var.environment}` GSA (`roles/container.defaultNodeServiceAccount` + `roles/artifactregistry.reader` only), `-compute@developer.gserviceaccount.com` is detached from all live `cage-staging` node pools and stripped to `roles/cloudtrace.agent` + `roles/logging.logWriter` on ``, and `scripts/bootstrap_gcp_project.sh` de-privileges the default Compute SA automatically on fresh bootstrap (`2026-11-14`).** Merged in `#658` (`38e12586`, [`infra/modules/gcp_gke_cluster/main.tf`](../infra/modules/gcp_gke_cluster/main.tf)) and `#661` (`c58b0913`, [`infra/targets/gcp-gke/wif.tf`](../infra/targets/gcp-gke/wif.tf)), exercised in clean-room Gate W `#738` (`84f39e21`, certifying `8956f505` in [`compliance/gate_w_certification.json`](../compliance/gate_w_certification.json): `227` image pulls / `115` cold pulls / `0` container restarts on GKE 1.35 under those two node-GSA roles alone), verified live on `` (`2026-11-14`: `cage-staging` `RUNNING` on `1.36.4-gke.2377000` in `cage-vpc-staging` / `cage-subnet-staging` runs `primary-node-pool`, `clickhouse-node-pool` hosting ClickHouse for operational telemetry only and not audit custody, `general-spot`, and `gpu-node-pool-nvidia-l4` under `cage-gke-nodes-staging@.iam.gserviceaccount.com` with zero `default` pools; `-compute@developer.gserviceaccount.com` absent from `sa-cage-deployer` and `cage-gateway` Token Creator policies and revoked from `roles/storage.admin`, `roles/storage.objectViewer`, `roles/artifactregistry.createOnPushWriter`, `roles/artifactregistry.writer`, `roles/container.defaultNodeServiceAccount`, and `roles/aiplatform.user`), and guarded by [`tests/infrastructure/test_gke_node_pool_service_account.py`](../tests/infrastructure/test_gke_node_pool_service_account.py), [`infra/targets/gcp-gke/tests/w6.tftest.hcl`](../infra/targets/gcp-gke/tests/w6.tftest.hcl), [`tests/infrastructure/test_wif_deployer_isolation.py`](../tests/infrastructure/test_wif_deployer_isolation.py), and [`tests/infrastructure/test_gcp_project_bootstrap.py`](../tests/infrastructure/test_gcp_project_bootstrap.py). | 2026-11-14 | +| POAM-2026-221 | CA-2 / CA-7 / CM-8 / RA-5 / SA-11 | ✅ **CLOSED — Vacuous `pip install` CI gates (`stpa-freshness-check`, `sbom-generate`, `sbom-generation`) and `raise_on_error=False` / `ContentCheck.UNAVAILABLE` silent fallback eliminated, Python SBOM floor tied to `uv.lock`, committed 8-package scratch SBOM retired, and all STPA & SBOM citations audited and qualified.** Merged in `#768` (`b92a1719`, `2026-11-15`; green mainline `build` run `43776695940` and `Security Scanning` run `43776695949`): [`render_default_oscal_artifacts()`](../src/gateway/governance/oscal_ssp_exporter.py) defaults `raise_on_error=True`, [`check_stpa_freshness.py`](../scripts/check_stpa_freshness.py) removes `ContentCheck.UNAVAILABLE` / commit-order fallback (red-before-green verified in Actions run `43765572946` / job `132969776429`), [`generate_sbom.py`](../scripts/generate_sbom.py) and `ci.yml::sbom-generate` enforce `expected_uv_lock_component_floor()` (`>= 90%` of platform-applicable `uv.lock` packages), all six workflow jobs run locked `uv sync`, `9b2b745d:compliance/sbom/python-deps-2026-06-08.cdx.json` + `9b2b745d:compliance/sbom/SBOM_SUMMARY.md` are deleted (`0` external `:`/`^:` pins across the repo), Gate W (`8956f505` / `fa772122`), all 9 Releases, GHCR, and all 209 census rows + `US_FED` SSP + Lula specs were swept (`0` `HOLDS`/`implemented` claims affected), and prose claims in `POAM-2026-176`, [`POAM_US_FED.md`](compliance/us_fed/POAM_US_FED.md) (`POAM-010`), [`NIST_AI_600_1_US_FED_ANALYSIS.md`](compliance/us_fed/NIST_AI_600_1_US_FED_ANALYSIS.md), [`THRESHOLD_TRACEABILITY_MATRIX.md`](../compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md), [`NIST_RMF_CHUNK5_MONITOR_ROADMAP.md`](compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md), and [`ISO42001_MANAGEMENT_REVIEW.md`](compliance/universal/ISO42001_MANAGEMENT_REVIEW.md) (dated addendum) are qualified. Guarded by [`tests/test_stpa_freshness_gate.py`](../tests/test_stpa_freshness_gate.py) and [`tests/unit/test_ci_workflow_dependency_sync.py`](../tests/unit/test_ci_workflow_dependency_sync.py) (detail section). Lula: not applicable (CI workflow and gate scripts; no Kubernetes resource changed). | 2026-11-15 | --- @@ -4511,9 +4511,10 @@ The scan also matched two entries that name a non-Active manifest without restin **Control:** NIST CA-2 (Control Assessments), CA-7 (Continuous Monitoring), CM-8 (System Component Inventory), RA-5 (Vulnerability Monitoring and Scanning), SA-11 (Developer Testing and Evaluation) **Risk Level:** High -**Status:** Open — remediated in `#768` (`2026-11-15`); held Open pending mainline squash-merge SHA +**Status:** Closed (2026-11-15) — remediated and merged on `main` in `#768` (`b92a1719`, branch head `b1b2ca14`), with prose citations qualified across `compliance/` and `docs/compliance/` **Date Opened:** 2026-11-15 **Target Closure:** 2026-11-30 +**Date Closed:** 2026-11-15 **Description:** Making [`scripts/check_stpa_freshness.py`](../scripts/check_stpa_freshness.py) fail closed in `#768` (`e3e253b9`, `2026-11-15`, Actions run `43763637543` / job `132963751680`) immediately failed `STPA Artifact Freshness Check` (`ci.yml::stpa-freshness-check`) with `ModuleNotFoundError: No module named 'opentelemetry'`, exposing that `stpa-freshness-check` had been silently skipping all three `US_FED` OSCAL content checks in GitHub Actions while reporting green. A repository-wide sweep of all `.github/workflows/*.yml` jobs for hand-listed `pip install` steps without `uv sync` uncovered two additional vacuous Python SBOM gates of the identical shape (`ci.yml::sbom-generate` and `security-scan.yml::sbom-generation`) plus three Lula validation jobs (`ci.yml::lula-ai600-validation`, `lula-validate.yml::lula-validate`, `lula2-pr-compliance.yml::lula2-crawl`) carrying fragile hand-listed `pip install pyyaml` setups. @@ -4526,23 +4527,26 @@ Making [`scripts/check_stpa_freshness.py`](../scripts/check_stpa_freshness.py) f | `ci.yml::sbom-generate` | `#16` (`8c6e9361`, `2026-06-15`) | `8c6e9361`..`9b2b745d` (`2026-06-15` to `2026-11-15`, **5 months**) | Installed only `pip install cyclonedx-py` without `uv sync` and checked only `d.get('bomFormat') == 'CycloneDX'`, generating and uploading `sbom.json` with **zero CAGE dependencies** | Container-image CycloneDX SBOMs in [`sbom.yml`](../.github/workflows/sbom.yml) (`trivy image --format cyclonedx`) | | `security-scan.yml::sbom-generation` | `1902c92c` (`2026-06-01`, `v2.0.0-dev.1`) | `1902c92c`..`9b2b745d` (`2026-06-01` to `2026-11-15`, **5.5 months**) | Installed only `pip install cyclonedx-bom` without `uv sync`, so [`scripts/generate_sbom.py`](../scripts/generate_sbom.py) (`--type python`) inventoried an empty runner environment (`cyclonedx-bom` + `pip`, **zero CAGE dependencies**) and `grype` scanned that empty Python SBOM | `security-scan.yml::python-dependency-audit` ran `uv sync --frozen --all-groups --all-extras` + `pip-audit`, and `sbom-generation` also ran `syft dir:.` + `Trivy` | -**Citation, Gate W Certification, and External Publication Audit (`2026-11-15`):** +**Citation, Census, Gate W Certification, and External Publication Audit (`2026-11-15`):** 1. **Closed `POAM-2026-176` (`docs/POAM.md`):** Cited `check_stpa_freshness.py` checking all 3 `US_FED` OSCAL artifacts in CI. Qualified with a dated correction (`2026-11-15`, `POAM-2026-221`): the closure remains valid on `test_stpa_freshness_gate.py` (`pytest-logic`) and `test_oscal_ssp_exporter.py`. 2. **Closed `POAM-010` ([`docs/compliance/us_fed/POAM_US_FED.md`](compliance/us_fed/POAM_US_FED.md)):** Cited `Grype, and CycloneDX SBOM generation` in `security-scan.yml`. Struck and qualified (`2026-11-15`, `POAM-2026-221`): `POAM-010`'s Python vulnerability coverage stood on `python-dependency-audit` (`pip-audit` over `uv sync --frozen`) and Trivy/Syft, not the vacuous Python `CycloneDX SBOM` leg. 3. **[`COMPLIANCE.md`](../COMPLIANCE.md) (`line 259`):** Programmatic `oscal_ssp_exporter` freshness paragraph annotated with `POAM-2026-221`. -4. **OSCAL `sa-11` / Component Remarks & Claim Census:** Zero Lula manifests cite `stpa-freshness-check`. Ten rows in [`compliance/oscal/evidence_claim_census.jsonl`](../compliance/oscal/evidence_claim_census.jsonl) mention `ci:stpa-freshness-check`; every one of those ten rows was already graded `WEAKER` or `ROLLUP` (`disposition: "restate"`), so zero `HOLDS` claims rested on it. +4. **OSCAL Claim Census (`209` rows), Lula (`33` manifests), and `US_FED` SSP — both STPA and SBOM/CycloneDX/Grype swept (`0` `HOLDS` or `implemented` claims affected):** + - **STPA side:** Zero Lula manifests cite `stpa-freshness-check`. Ten rows in [`compliance/oscal/evidence_claim_census.jsonl`](../compliance/oscal/evidence_claim_census.jsonl) mention `ci:stpa-freshness-check`; all ten were already graded `WEAKER` or `ROLLUP` (`disposition: "restate"`). + - **SBOM / CycloneDX / Grype side:** Swept `sbom`, `cyclonedx`, `grype`, `syft`, `security-scan`, `generate_sbom`, `cm-8`, `ra-5`, `sa-12`, and `sr-3` across all `209` census entries, `compliance/lula/**`, and [`compliance/oscal/system-security-plan.yaml`](../compliance/oscal/system-security-plan.yaml). Census rows `19` & `20` (`ra-5`) are `planned` (`NOT_A_CLAIM`), row `39` (`si-3` platform) is `NOT_IN_FORCE`, and row `209` (`common-controls-catalog.yaml` `si-3` shared) had already graded its customer leg `customer_verdict: "WEAKER"` (`disposition: "fix"`). Only Stub-rated [`compliance/lula/lula-validation-ra5.yaml`](../compliance/lula/lula-validation-ra5.yaml) mentions `security-scan.yml`/`sbom.yml` in `Notes`, and `US_FED` SSP `ra-5` (`lines 1336–1365`) is `planned` and cites `pip-audit, Trivy SBOM/CVE scanning, and OPA policy tests`. + - **Prose claims qualified (`2026-11-15`):** Corrected [`compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md`](../compliance/risk_acceptance/THRESHOLD_TRACEABILITY_MATRIX.md) (`THR-AUD-003`, `CM-8`, `RA-5`, and `§5.3 RA-003`, withdrawing both its false K8s SBOM `CronJob` / `24h` container-scan claim under Open `POAM-2026-010` and its unqualified `sbom-generation` compensating-control claim), qualified `P2-4` in [`docs/compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md`](compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md), and appended a dated `2026-11-15` addendum to [`docs/compliance/universal/ISO42001_MANAGEMENT_REVIEW.md`](compliance/universal/ISO42001_MANAGEMENT_REVIEW.md) (`§2.1`) preserving the `2026-06-23` draft record verbatim. 5. **Gate W Certification Audit (`8956f505` `2026-11-13` and `fa772122` `v4.1.0` — NEGATIVE, zero reliance):** Certified Gate W fell inside all three vacuous windows (`2026-11-13` vs. `9b2b745d` on `2026-11-15`). Audited [`compliance/gate_w_certification.json`](../compliance/gate_w_certification.json) (all 184 lines covering `8956f505` and `fa772122`), the published GitHub Release WORM archive `gate-w-8956f505-clean-room-worm-archive.tar.gz` (all 58 tar entries + `manifest.json`), [`scripts/staging_lifecycle.sh`](../scripts/staging_lifecycle.sh), and [`scripts/run_staging_validation_once.sh`](../scripts/run_staging_validation_once.sh) (`lines 397, 403`, which invokes `python -m src.gateway.governance.oscal_ssp_exporter export` directly on the synced test runner rather than `stpa-freshness-check`). **Zero entries in Gate W's certification or published WORM archive cite `stpa-freshness-check`, `sbom-generate`, `sbom-generation`, `sbom`, or `cyclonedx`**; no certification qualification or public WORM archive addendum is required. 6. **External SBOM Publication & Committed Public-Repository Audit (`2026-11-15`):** - **Workflow-generated empty SBOMs stayed strictly internal:** Audited all 9 published GitHub Releases (`conformance-v0.4.1`, `v4.1.0`, `gate-w-8956f505`, `client-v0.3.0`, `v4.0.0`, `v3.3.0`, `v3.2.0`, `client-v0.2.0`, `client-v0.1.0` — zero SBOM assets or release-note claims), [`publish-client-sdk.yml`](../.github/workflows/publish-client-sdk.yml) (pushes wheels/sdists and GHCR `cage-client` / `cage-conformance` images with zero `cosign attest` or `actions/attest-sbom` steps), GitHub Attestations API (`404` — zero attestations published), repository Actions secrets (`SBOM_S3_BUCKET` / `SBOM_S3_ACCESS_KEY` unset, so [`sbom.yml`](../.github/workflows/sbom.yml)'s S3/GCS upload step was always skipped), and `docs/partners/` (zero SBOM claims). - - **Committed public-repository 8-package scratch SBOM discovered and retired:** `9b2b745d:compliance/sbom/python-deps-2026-06-08.cdx.json` and `9b2b745d:compliance/sbom/SBOM_SUMMARY.md` had been committed directly to the public repository on `2026-06-08` from an unsynced gRPC scratch environment (`generate_sbom.py/pip-fallback`) containing **only 8 bootstrap packages** (`certifi`, `grpcio`, `grpcio-tools`, `pip`, `protobuf`, `setuptools`, `typing_extensions`, `wheel` — zero CAGE runtime dependencies), and were cited at `docs/compliance/us_fed/NIST_AI_600_1_US_FED_ANALYSIS.md` (`line 544`). `#768` deletes both committed files (`git rm`), strikes and withdraws the `NIST_AI_600_1_US_FED_ANALYSIS.md` citation, and updates [`compliance/sbom/README.md`](../compliance/sbom/README.md) so static SBOM snapshots are not committed to git where they drift from `uv.lock`. + - **Committed public-repository 8-package scratch SBOM discovered, retired, and pin-swept:** `9b2b745d:compliance/sbom/python-deps-2026-06-08.cdx.json` and `9b2b745d:compliance/sbom/SBOM_SUMMARY.md` had been committed directly to the public repository on `2026-06-08` from an unsynced gRPC scratch environment (`generate_sbom.py/pip-fallback`) containing **only 8 bootstrap packages** (`certifi`, `grpcio`, `grpcio-tools`, `pip`, `protobuf`, `setuptools`, `typing_extensions`, `wheel` — zero CAGE runtime dependencies), and were cited at `docs/compliance/us_fed/NIST_AI_600_1_US_FED_ANALYSIS.md` (`line 544`). `#768` (`b92a1719`) deleted both committed files (`git rm`), struck and withdrew the `NIST_AI_600_1_US_FED_ANALYSIS.md` citation, and updated [`compliance/sbom/README.md`](../compliance/sbom/README.md); a repository-wide grep for `python-deps-`, `SBOM_SUMMARY`, and `[0-9a-f]{7,40}\^?:compliance/sbom` confirmed **zero external `:` or `^:` pins** resolve to either retired file. **Remediation (`2026-11-15`):** 1. **Inability-to-check fails closed:** [`render_default_oscal_artifacts()`](../src/gateway/governance/oscal_ssp_exporter.py) defaults `raise_on_error=True` and rejects empty target renders; [`scripts/check_stpa_freshness.py`](../scripts/check_stpa_freshness.py) deletes `ContentCheck.UNAVAILABLE` and the entire `_check_order` / `_check_commit_order` / `_check_working_tree` fallback, failing with exit code `1` on any import error, missing `ruff` binary on formatted targets, unregistered artifact, missing file, `None` render, or SHA-256 mismatch. -2. **Dynamic `uv.lock`-derived Python SBOM component floor:** [`scripts/generate_sbom.py`](../scripts/generate_sbom.py) (`expected_uv_lock_component_floor()`, computing `>= 90%` of platform-applicable non-workspace `[[package]]` entries in `uv.lock`, currently `>= 276` components) and `ci.yml::sbom-generate` fail closed on both empty venvs and partial/half-rendered SBOMs, and invoke `cyclonedx_py environment` cleanly under `cyclonedx-bom` v7. +2. **Dynamic `uv.lock`-derived Python SBOM component floor:** [`scripts/generate_sbom.py`](../scripts/generate_sbom.py) (`expected_uv_lock_component_floor()`, computing `>= 90%` of platform-applicable non-workspace `[[package]]` entries in `uv.lock`, currently `>= 276` components) and `ci.yml::sbom-generate` fail closed on both empty venvs and partial/half-rendered SBOMs, and invoke `cyclonedx_py environment` cleanly under `cyclonedx-bom` v7 (`_extract_sbom_tool_name()`). 3. **Locked `uv sync` across CI workflows:** [`ci.yml`](../.github/workflows/ci.yml) (`stpa-freshness-check`, `lula-ai600-validation`, `sbom-generate`), [`security-scan.yml`](../.github/workflows/security-scan.yml) (`sbom-generation`), [`lula-validate.yml`](../.github/workflows/lula-validate.yml) (`lula-validate`), and [`lula2-pr-compliance.yml`](../.github/workflows/lula2-pr-compliance.yml) (`lula2-crawl`) replace hand-listed `pip install` steps with `astral-sh/setup-uv` + `uv sync --locked` / `--frozen`. 4. **Structural regression guards:** [`tests/test_stpa_freshness_gate.py`](../tests/test_stpa_freshness_gate.py) (replaying the `opentelemetry` `ModuleNotFoundError`, missing `ruff`, un-reciped artifacts, and planted STPA/SSP content drift) and [`tests/unit/test_ci_workflow_dependency_sync.py`](../tests/unit/test_ci_workflow_dependency_sync.py) (verifying AST stdlib purity on pre-`uv` scripts, requiring `uv sync` on every workflow job running repository scripts, and rejecting both empty 1-component and partial 150-component Python SBOMs against `expected_uv_lock_component_floor()`), both binding `POAM-2026-221`. **Closure Criteria:** 1. ✅ **Verified (`2026-11-15`, `#768` `6f04ba7d`):** Red-before-green confirmed live in GitHub Actions (`run 43765572946`, `STPA Artifact Freshness Check` job `132969776429`): planted drift in `config/rails/generated_stpa_rails.co` failed `check_stpa_freshness.py` with `STALE (content differs from a fresh compile (sha256)): config/rails/generated_stpa_rails.co`, planted drift in `compliance/oscal/system-security-plan.yaml` failed `check_stpa_freshness.py` with `STALE (content differs from a fresh compile (sha256)): compliance/oscal/system-security-plan.yaml`, and the restored checkout passed all 9 artifacts (`9/9 content match`). In the same push, `ci.yml::sbom-generate` (`job 132969811995`) validated `CycloneDX v1.6 (323 components)` and `security-scan.yml::sbom-generation` (`run 43765572949`, `job 132969955610`) built and Grype-scanned `323 components` (`0 vulnerabilities`). -2. Merged to `main` with the mainline squash-merge SHA recorded here, and [`tests/test_stpa_freshness_gate.py`](../tests/test_stpa_freshness_gate.py) and [`tests/unit/test_ci_workflow_dependency_sync.py`](../tests/unit/test_ci_workflow_dependency_sync.py) passing on `main`. +2. ✅ **Verified (`2026-11-15`, `#768` squash-merged on `main` at `b92a1719`):** Mainline `build` run `43776695940` (including [`tests/test_stpa_freshness_gate.py`](../tests/test_stpa_freshness_gate.py), [`tests/unit/test_ci_workflow_dependency_sync.py`](../tests/unit/test_ci_workflow_dependency_sync.py), `stpa-freshness-check`, and `sbom-generate`) and mainline `Security Scanning` run `43776695949` (`sbom-generation` with `expected_uv_lock_component_floor()` and CycloneDX 1.6 summary report) passed on `b92a1719`. diff --git a/docs/compliance/universal/ISO42001_MANAGEMENT_REVIEW.md b/docs/compliance/universal/ISO42001_MANAGEMENT_REVIEW.md index c6413d078..476b82ce6 100644 --- a/docs/compliance/universal/ISO42001_MANAGEMENT_REVIEW.md +++ b/docs/compliance/universal/ISO42001_MANAGEMENT_REVIEW.md @@ -34,6 +34,8 @@ Per ISO 42001 §9.3.2, the following inputs were reviewed: - POAM-020 (CM-3): Technical report README version mismatch corrected - POAM-021 (SI-4): AgentSight remote exporter confirmed already configured +> **Note (2026-11-15):** On the `2026-06-23` review date (inside the `1902c92c` `2026-06-01` .. `9b2b745d` `2026-11-15` window, `POAM-2026-221` in [`docs/POAM.md`](../../POAM.md)), the Python CycloneDX + Grype leg of `.github/workflows/security-scan.yml` (`sbom-generation`) ran `pip install cyclonedx-bom` without `uv sync`, inventorying and Grype-scanning zero CAGE Python dependencies. Python CVE coverage during Q2 2026 rested on `python-dependency-audit` (`uv sync --frozen --all-groups --all-extras` + `pip-audit`) and Trivy/Syft filesystem scanning rather than the Python Grype leg, which was restored in `#768` (`b92a1719`, `2026-11-15`) with `expected_uv_lock_component_floor()`; periodic in-cluster scanning remains open under `POAM-2026-010`. + ### 2.2 Lula Audit Results (Q2 2026) | Manifest | Result | Evidence | diff --git a/docs/compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md b/docs/compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md index 98edc9331..78c96e004 100644 --- a/docs/compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md +++ b/docs/compliance/us_fed/NIST_RMF_CHUNK5_MONITOR_ROADMAP.md @@ -335,7 +335,7 @@ SC-4 (OPA ConfigMap label compliance) has continuous drift detection via the 60- | P2-1 | `src/gateway/server/governance_middleware.py`, `de5534ac:deployment/k8s/backend-deployment.yaml.tpl` | Harden `CAGE_ROUTING_SEAL_SECRET` bypass path: add HMAC verification on routing override requests; log all bypass attempts as `GOVERNANCE_VIOLATION` events. Remove ` | | true` fallback. | SC-12, AC-3, AU-3 | 3d | | P2-2 | `compliance/lula/sp80053/lula-ac-3.yaml`, `lula-sc-8.yaml`, `lula-si-4.yaml`, `lula-ia-3.yaml` | Create 4 new Lula SP 800-53 validation manifests using Kubernetes domain: check OPA Deployment ready, NetworkPolicy count ≥ 9, AgentSight DaemonSet running, mTLS annotation present | AC-3, SC-8, SI-4, IA-3 | 5d | | P2-3 | `de5534ac:deployment/k8s/lula-cron.yaml` | Add `sp80053/` validations to CronJob `--input` flags; extend OSCAL Assessment Results to include SP 800-53 findings | CA-7, CA-2 | 2d | -| P2-4 | ~~`scripts/vulnerability_scan.sh` (new)~~ | ~~CI step: `trivy image --exit-code 1 --severity CRITICAL ` for all 5 container images; `pip-audit` for Python deps; output SARIF to GitHub Security tab~~ **✅ COMPLETED:** Implemented in `.github/workflows/security-scan.yml` — pip-audit, Trivy, Grype, CycloneDX SBOM. See POAM-010 (Closed). | RA-5, SI-2, CM-8 | ✅ | +| P2-4 | ~~`scripts/vulnerability_scan.sh` (new)~~ | ~~CI step: `trivy image --exit-code 1 --severity CRITICAL ` for all 5 container images; `pip-audit` for Python deps; output SARIF to GitHub Security tab~~ **✅ COMPLETED:** Implemented in `.github/workflows/security-scan.yml` — `pip-audit`, Trivy, Syft/Grype filesystem SBOM, and `.github/workflows/sbom.yml` (4 container images on PR/push). See POAM-010 (Closed). *(Correction `2026-11-15`, `POAM-2026-221`: from `1902c92c` `2026-06-01` through `9b2b745d` `2026-11-15`, the Python CycloneDX + Grype leg in `security-scan.yml::sbom-generation` ran without `uv sync` and scanned zero CAGE Python packages; Python CVE coverage stood on `pip-audit` until `#768` `b92a1719` restored `uv sync` + `expected_uv_lock_component_floor()`, and periodic in-cluster container scanning remains Open under `POAM-2026-010`.)* | RA-5, SI-2, CM-8 | ✅ | | P2-5 | `compliance/sbom/` (new directory) | Generate SBOM via `syft` or `cyclonedx-py` for all container images and Python packages; store in `compliance/sbom/-.json`; link from SSP | SA-12, SR-3, CM-8 | 3d | | P2-6 | `deployment/k8s/` + all vLLM manifests | Replace `:latest` image tags with `@sha256:` via `docker inspect --format={{.RepoDigests}} ` in deploy script; pin all images | CM-8, SI-2 | 2d | | P2-7 | `deployment/terraform/iam.tf` | Audit all Service Account roles; remove `roles/editor` or `roles/owner` if present; apply least-privilege IAM roles per service (`financial-advisor-sa`: `storage.objectViewer` + `secretmanager.secretAccessor` only) | AC-6, IA-5 | 3d |