SentinelOS
AI Safety Operating Layer
SentinelOS governs AI systems.
SentinelOS is a modular AI Safety Operating Layer composed of PROACTIVE Gov, HUI Guard, Eval Workbench, Red Team Lab, and Trace Console. This portfolio acts as the inspection console where reviewers can understand what exists today, how incidents are handled, and where each safety control lives.
Architecture Snapshot
User Interaction
│
▼
HUI Guard
│
▼
PROACTIVE Gov
│
▼
Eval Workbench
│
▼
Red Team Lab
│
▼
Trace ConsoleFor a full description of each module, see docs/SentinelOS_ARCHITECTURE.md.
Incident Lifecycle (High Level)
- AI system emits a confident claim about functionality or safety posture.
- PROACTIVE Gov verifies the claim against code, config, and evidence.
- Eval Workbench scores behavior against evaluation rubrics.
- HUI Guard assesses human impact and intervenes if necessary.
- Trace Console records the full evidence trail for later inspection.
See docs/SentinelOS_INCIDENT_LIFECYCLE.md for the detailed walkthrough.
SentinelOS Truth-Status
Every SentinelOS module and artifact is explicitly labeled so reviewers can see what exists today and what is planned.
Modules
Constitutional AI safety agent. Validates MR claims against I1–I6 invariants.
Loop detection and rescue for neurodivergent-friendly development.
Cognitive safety evaluator for learning environments.
Agent consent governance with cryptographic ledger and policy engine.
Human-UI governance layer for epistemic labeling in decision support.
Empirical Safety Engine — failure classification, safety scoring, trend analysis.
Artifacts
100% detection rate, 0% FPR on TruthfulQA. Evidence in docs/evidence/validation_results.json
docs/SentinelOS_ARCHITECTURE.md
docs/SentinelOS_INCIDENT_LIFECYCLE.md
docs/SentinelOS_TRUTH_STATUS.md
docs/SentinelOS_SAFETY_CASE.md
safety-identity-refactor/ — 36 files, Steps 0-19 of 25-step contract