What you get
Guardrails
- Content-free evidence — IDs, counts, categories, timestamps, and file patterns only. Never prompts, code, commands, or secret values.
- Detection is automatic; enforcement is human — revoke / rotate / dismiss decisions stay on an admin or reviewer.
- Previews before irreversible actions where the product already requires confirmation (for example key revoke).
- Legal hold — compliance retention never deletes under legal hold; lowering compliance retention warns on Control health (RC-AI-10).
Control catalog (shipped)
Controls that need later features (vendor register, incidents) are seeded when those features ship so Control health does not show noisy
na / fail rows early.
Status meanings
Access reviews
Admins create a campaign with a label, due date, and scope (gateway keys, org API keys, members). Items freeze a subject snapshot at creation. Team-scoped keys go to a team lead when possible; reviewers cannot decide their own access when another admin exists (single-admin self-review is allowed and recorded as an SoD exception → RC-AI-07 warn). Decisions arekeep / revoke / modify / rotate with a required reason. Completing a campaign requires attestation; the product stores completed_by, timestamp, and a SHA-256 of the decision set (no cryptographic signature).
Removing a member also creates off-cycle shared key exposure items for team gateway keys the departed member could use but did not create. Reviewers choose rotate or keep (with justification).
Risk exceptions
A single register covers kinds such asaccepted_finding, dlp_exception, and control_threshold_override. Every exception needs a rationale and an expires_at within 12 months of approval. On expiry, status becomes expired and reversible prior state is restored where applicable.
On Signals, Accept risk on an anomaly finding creates an accepted_finding exception and dismisses the finding.
Related pages
- Compliance dashboard — Control health, access reviews, exceptions, evidence export
- Coverage Alerts — device DLP and hook integrity feeding RC-AI-01 / 04
- Anomalies — detectors behind RC-AI-02 / 03 / 04
- Signals dashboard — triage and Accept risk
- Audit —
access_review.*andrisk_exception.*events