MAQPNA

Learn

The AI agent audit trail: tamper-evident and verifiable.

A log tells you what happened. A tamper-evident log lets someone else prove it.

Updated 31 October 2026.

FAQ

Common questions

Is application logging enough to audit an AI agent?

Usually not. Application logs are written by the same process the agent runs in, can be edited, and rarely record the policy decision. An audit trail written outside the agent, with hash chaining and signatures, is evidence rather than a log.

What is the difference between tamper-proof and tamper-evident?

Tamper-proof means a record cannot be changed. Tamper-evident means any change can be detected. Hash chains and signatures give tamper-evidence; write-once storage adds protection against deletion.

Should an audit trail store the agent's prompts?

Storing a hash of the arguments is often safer. It proves what was sent without keeping personal data or secrets in the audit system. Keep full content, if you need it, in a separate store with its own access controls.

How does an auditor verify a MAQPNA ledger?

They run maqpna audit verify on a copy of the ledger with the public keys from the gateway's JWKS endpoint. It recomputes the hash chain and checks every signed checkpoint, offline, with no secret needed.

Evidence you can hand to an auditor.

Run an agent under a local MAQPNA, then verify its ledger with one command.

More install options
curl -fsSL https://maqpna.com/install.sh | sh