QED Proof
Start free
Menu

AI agent verification

Independently confirming that an AI agent's claimed work actually happened, using evidence the agent didn't produce — and making the result checkable by others.

Also called: agent verification, verifying AI agents

In plain words

AI agent verification is answering one question from outside the agent: did the work it says it did really happen? It isn't evaluating whether the agent is smart, or reviewing its code, or scoring its reasoning. It's checking outcomes.

Independence is the whole point. If the evidence comes from the agent, or from a log written inside the agent, you've only re-read its report. Real verification uses evidence the agent couldn't have produced: the state of the destination.

How QED Proof uses it

QED Proof is an independent verifier. Each check runs the same sequence:

  1. An agent claim arrives, untrusted.
  2. A verifier reads the destination over read-only access.
  3. A verdict is decided deterministically.
  4. The result becomes a signed receipt.
  5. The receipt joins an append-only log.
  6. The log is anchored on-chain.

The last three steps make the verification itself verifiable, so a third party doesn't have to trust QED Proof either. QED Proof is deliberately not a code reviewer: a check like "CI passed on that commit" is one verifier among many, and it never judges code quality.

Example

Coding and ops agents are verified today, against GitHub and public URLs, and so are agents that post to X or Slack. Meta (Pages, Instagram, Ads) connects read-only now, with its verifiers next. See use cases.