// AI governance
AI governance
& decision proof.
AI systems that make consequential decisions need evidence that goes beyond internal logs. Attesto records model versions, prompt and output hashes, policy checks, and human oversight events as cryptographic proof objects — outside the AI system itself.
Model version proof
Record the exact model version, weights reference, and configuration used for each inference event — immutably tied to the decision timestamp.
Prompt and output hashes
Hash input prompts and AI outputs at inference time. Any later claim about what the model was asked or answered can be verified against the original hash.
Policy check evidence
Capture guardrail checks, content filtering events, and safety policy evaluations as tamper-evident proof — showing that compliance controls were active.
Human oversight records
Prove that human review, approval, or override events occurred for high-risk decisions — with timestamps and signed receipts that cannot be rewritten.
Bias and accuracy monitoring
Record model monitoring events — performance metrics, drift detection, fairness checks — as anchored audit records for post-deployment review.
Incident trace
When an AI decision is challenged, reconstruct the full decision context from cryptographic evidence without relying on the AI system to report its own behaviour.