Non-Repudiation
The property that a party cannot credibly deny having taken an action, because the evidence is strong enough to hold up even when they'd very much like to say 'that wasn't us.' For AI agents it cuts two ways. Externally, it lets you *prove* to a customer, auditor, or court exactly what your agent did and didn't do. Internally, it stops your own operators — or a compromised insider — from rewriting the record after a bad outcome. In practice it means cryptographically signed, tamper-evident logs: actions signed by the agent's identity key, audit-trail entries chained by hash, timestamps from a trusted source, and storage the operator cannot silently alter (append-only, write-once, or externally witnessed). Without non-repudiation an audit trail is only as trustworthy as the party holding it — which is exactly the party with the strongest motive to edit it after an incident. It is the technical backbone that makes accountability survive a dispute.
Written by Max Zeshut
Founder at Agentmelt · Last updated Sep 9, 2026
Example
A fintech's payment agent signs every action with its own identity key and chains each log entry to the previous one's hash. Six months later a customer disputes a transfer. The company produces a signed, unbroken hash chain showing the exact action, its authority, and the human approval at the gate — evidence the customer, and a regulator, can verify was not altered after the fact.
Frequently asked questions
- Why isn't a normal database log enough for non-repudiation?
- Because whoever controls the database can edit it. A plain log proves what happened only if you trust the party holding it — and after an incident, that party has the strongest motive to change the story. Non-repudiation removes that trust requirement: signing and hash-chaining make any after-the-fact edit detectable, so the record stands on its own.
- Is non-repudiation overkill for low-risk agents?
- Often, yes. A read-only research agent rarely needs signed, chained logs. Reserve full non-repudiation for agents that touch money, regulated data, legal actions, or anything you might have to *prove* later. Match the strength of the evidence to the stakes of the action — the same tiering you use for approval gates and scopes.