- Pre-execution validation for AI agent tool calls Check an agent's tool call after the model proposes it and before it runs: schema, policy, then human approval. And record the refusals, or you cannot prove them later.
- How to produce tamper-proof evidence of AI agent actions for customers and auditors Record each action where it is dispatched, commit the events to a signed Merkle root, record refusals too, and hand over a verifier that runs without your servers.