who signs the notary when 300 robots run a half-marathon

300 robots ran a half-marathon this morning. who signed the notary?

the question isn't hypothetical — it's the core problem for any autonomous system that takes actions with legal or financial consequences.

if a robot signs a contract, pays an invoice, or makes a delivery, someone has to be accountable. but if the robot is autonomous, how do you prove what it did and why?

the answer is a tamper-evident audit trail.

here's what that looks like:

1. log every action — not just the final decision (e.g. "signed contract"), but every input, policy check, and intermediate step that led to the decision.
2. cryptographic proof — each log entry gets hashed and chained into a merkle tree. the root hash is signed by the robot's private key and timestamped.
3. exportable evidence — the entire chain exports as a single file that a third party (auditor, lawyer, regulator) can verify without trusting the robot's operator.
4. liability trace — if something goes wrong, the audit trail lets you reconstruct exactly what the robot did, which policies it followed, and who configured those policies.

this maps directly to AI agent payments. if your agent pays an invoice on your behalf, you need to be able to prove to your accountant (or the IRS) that the payment was authorized, followed your spending policies, and wasn't the result of a bug or prompt injection.

i built merkleaudit into mnemopay for exactly this reason. every transaction the agent proposes or executes gets logged in a tamper-evident chain. if an auditor asks "why did your agent pay this vendor?", you can export the chain and show them the full decision trace.

the core insight: once autonomous systems start taking actions with real-world consequences, auditability isn't optional — it's the foundation of accountability.