Protect the control plane.
Who watches the control plane? Imperium governs every AI. Sentinel governs Imperium. It runs outside the main trust domain, with its own baseline, its own evidence store and its own channel to your security team.
WHAT SENTINEL WATCHES
- Binary integrity — policy engine, gateway, connectors: expected hash vs observed hash, continuously.
- Policy integrity — signed policy bundles; an unsigned or altered rule set is never enforced.
- Admin sessions — privileged actions verified against the baseline; frozen on drift.
- Evidence store — append-only, sealed, verifiable after the fact.
- Runtime sensors — file integrity, process lineage, network egress of the appliance itself.
AN INCIDENT
POLICY ENGINE BINARY MODIFIED. Expected 8DB2A711…, observed 4FF87A03…. Sentinel acts: FREEZE ADMIN SESSION → SEAL EVIDENCE → ENTER SAFE ENFORCEMENT → NOTIFY SECURITY. Imperium keeps enforcing the last known-good policy while the incident is handled. Then: CONTROL RESTORED.
SAFE ENFORCEMENT
When trust in the control plane itself is in question, Sentinel does not switch AI off and it does not open the gates. It pins the last verified policy, denies anything that would require a fresh admin decision, and keeps the evidence flowing until an operator verifies the system.