Federal posture
Every claim on this page links to a source artifact in the GRIFF AI repository. Where a control is not yet in place we say so out loud, with the audit ID that tracks the gap. Federal procurement teams: this page is intended to short-circuit the first round of security questionnaires.
section-889-attestation/v1
A CI-blocking scanner walks every commit looking for references to covered vendors enumerated by NDAA 2019 §889 / FAR 52.204-25 / 2 CFR 200.216: Huawei, ZTE, Hytera, Hikvision, Dahua. It scans package.json, pyproject.toml, go.mod, Dockerfile*, CI workflow YAML, and source trees. A non-zero exit blocks merge. Attestation JSON is emitted on every run with schema id section-889-attestation/v1.
Source: A:\projects\v2-platform\src\compliance\section_889.py · Vendor registry: covered_vendors_section_889.yaml
Caveats published with the scanner: no transitive dependency resolution, no image-layer inspection; operator ratification required before treating output as a legal compliance assertion.
Each row links to the in-repo artifact that is the primary evidence today. These are coverage statements against specific controls, not a claim of FedRAMP authorization (see “what we do not claim” below).
| Control | Name | Status | Evidence + source |
|---|---|---|---|
| AC-2(3) | Service-token rotation (30-day lifetime) | In place | Cloudflare Access service tokens rotated on a 30-day cycle. Closes the V-8 finding that 1-year token lifetime was ~12x non-compliant with FedRAMP Moderate AC-2(3). Rotation runbook + bearer rotation already executed against memory.griff.run. A:\GRIFF_AI\01_INTERNAL\_audit\2026-05-10-final-drain-V8-claude-cf-posture.md (§1 row B2, §1 row E2) |
| AU-2 | Auditable events | In place | Every Intent Compiler / Autonomy-core mutation lands in the D1 ledger with actor, action, target, and timestamp. RunReplay rows record actor identity per request. A:\projects\v2-platform\src\autonomy-core\ledger\d1_ledger.py |
| AU-3 | Content of audit records | In place | Each audit record carries: actor id, action, target id, timestamp, prior-hash, entry-hash. Recall responses additionally carry sha256, source_path, and rank_reason (see sample below). A:\projects\v2-platform\src\autonomy-core\autonomy_layer_schemas.py (RunReplay model — Security Committee P1 #6 tamper-evident chain) |
| AU-12 | Audit generation — hash-chained | In place | RunReplay rows include prev_hash (SHA-256 of previous row’s canonical encoding) and entry_hash (SHA-256 of current row including prev_hash). Single-writer chain head; tampering with any row breaks the chain on the next verify. A:\projects\v2-platform\src\autonomy-core\autonomy_layer_schemas.py (lines 64–76) |
The point of EO 14028 §4 is provenance and traceability. We list what is in place, what is deliberately deferred, and what does not yet exist. Federal procurement reviewers should read this as the no-fluff version — if a row says “deferred” or “not yet” that is the honest answer.
Hash-chained RunReplay rows with single-writer chain head. Closes EO 14028 §4(e) tamper-evidence intent for our internal action ledger.
A:\projects\v2-platform\src\autonomy-core\autonomy_layer_schemas.py
CI-blocking scanner walks dependency manifests, Dockerfiles, CI YAML, and source trees for Huawei / ZTE / Hytera / Hikvision / Dahua references. Emits attestation JSON with schema-id section-889-attestation/v1. Exit 0 = clean, 1 = findings, 2 = scanner error.
A:\projects\v2-platform\src\compliance\section_889.py
cloudflared is auto-updated weekly without an in-band SHA-256 pin against the published Cloudflare digest. Tracked as a transparent gap in V-8 finding C4 (escalated to P1 under federal-AV lens). Workstream open; not yet shipped.
A:\GRIFF_AI\01_INTERNAL\_audit\2026-05-10-final-drain-V8-claude-cf-posture.md (§1 row C4)
No SBOM publication pipeline today. Roadmap item; not currently asserted. Section 889 scanner is the partial substitute for covered-vendor risk; it is not an SBOM.
— (gap; tracked for post-procurement engagement)
Every recall response includes a SHA-256 of the source document, the canonical source_path, and a rank_reason string that names the matcher. The response also carries the current audit chain head so the caller can independently verify chain integrity at the moment of the query.
{
"query": "How did we decide the 30-day token rotation cadence?",
"hits": [
{
"id": "RV4-token-rotation-decision-2026-05-10",
"score": 0.847,
"sha256": "8a4f7d2c91b6e3a5fd4c7e8b1a2d9f3e6c5b4a7d8e9f0a1b2c3d4e5f6a7b8c9d",
"source_path": "A:/GRIFF_AI/01_INTERNAL/_audit/2026-05-10-final-drain-V8-claude-cf-posture.md",
"rank_reason": "exact-phrase match on 'service-token rotation' + AC-2(3) FedRAMP context boost",
"snippet": "Service-token 1-year duration | P0 | UPHOLD | Single largest unaddressed blast-radius. FedRAMP Moderate AC-2(3) requires rotation; 1-year is non-compliant by ~12x."
}
],
"chain_head": {
"row_id": "runreplay-2026-05-10T18:42:11Z-7c91",
"entry_hash": "f3e2d1c0b9a8f7e6d5c4b3a2918f7e6d5c4b3a2918f7e6d5c4b3a2918f7e6d5c"
}
}Response shape source: A:\projects\v2-platform\src\autonomy-core\autonomy_layer_schemas.py
Published before procurement asks. These are not omissions of intent — they are statements that the work has not yet been done. Path to each is a procurement-conditional roadmap item.
No ATO, no 3PAO assessment in progress, not on the FedRAMP Marketplace. Path: pursue Low or Moderate baseline once a federal customer is signed; SaaS-on-Cloudflare boundary already maps cleanly to FedRAMP-authorized hyperscaler infrastructure where applicable.
No CMMC Level 1, 2, or 3 certification today. Path: pursue Level 2 in parallel with the first DoD-adjacent engagement; current control posture (AC-2(3), AU-2/3/12, hash-chained audit) was designed to be assessment-ready.
No Authorization to Operate against any federal system boundary. Path: stand up a customer-specific ATO package on first procurement engagement; the controls listed above are the starting evidence basket.
Contact
Email lands directly with the operator. Replies are typically within one business day. Under NDA we can share the closed-finding audit trail referenced on this page (V-3, V-6, V-8 cross-vendor reviews) plus the Section 889 attestation JSON from the most recent CI run.