The black box recorder for regulated AI agents captures who authorized an action, what it decided, and whether it matched policy, as tamper-evident, hash-linked records. Query the trail, simulate tampering, and export an examiner-ready evidence package.
-
Reset and prepare
Action Click Reset demo to start from a clean, freshly seeded chain.
-
Explore the seeded chain
Action Review the initial entries (credit decision, KYC escalation, blocked over-limit refund).
Concept This is the black box recorder for your AI agent. It shows who authorized the action, its context, the decision, and the policy match.
-
Seal a new record
Action Select a preset (Credit approval, KYC escalation or Policy breach) and click Seal into ledger.
Concept In production, a runtime hook inside the agent loop automates this step.
-
Filter and examine
Action Use the query filters: set decision to block or policy match to breached.
Concept This is the examiner view. Pull breaches or escalations instantly, without an engineering request.
-
Verify integrity
Action Click Verify integrity and confirm the status reads intact.
Concept Each block's cryptographic hash covers the previous block's hash, so the chain checks itself.
-
Test tampering
Action Click Simulate tampering. The edited block and every block after it turn red.
Concept Quietly editing history breaks every block after it.
-
Export and wrap up
Action Click Export evidence package, then click Reset demo to finish.
Concept Auditors get a verifiable bundle. True production independence also needs external anchoring, such as WORM storage or signed checkpoints.