How do I prove a human approved an AI agent's irreversible action?

With sebbi.pro Point of No Return, an irreversible agent action that has no way back is sent to a named person through a private link. The person sees the agent, the action and what it affects, types their name and reason, and approves or refuses. The gate seals a fingerprint of their name and reason, the moment of the decision and the seconds they spent on the screen before deciding; an approval given in under five seconds is flagged. The agent is only cleared after that sealed approval, and the public record at https://sebbi.pro/no-return/<code> lets an auditor, insurer or regulator check the sequence: asked, approved, carried out. This is the per-action evidence of human oversight that EU AI Act Article 14 describes: a person able to intervene, override and stop.

How it works

1. The agent calls POST /x/noreturn/check; with no way back the answer is NEEDS_HUMAN and a private approval link.
2. The person opens the link, reads the action, and approves or refuses with their name and reason.
3. The agent calls POST /x/noreturn/release and acts only on GO.
4. Share the public record at https://sebbi.pro/no-return/<code>.