EVIDENCE AND AUDIT

A decision should be explainable after the agent acts.

Exact-action authority evidence records what was proposed, whose limits applied, why the action was denied or authorised and what the recipient checked. It is designed to support review without turning a broad identity claim into blanket permission.

THE PROBLEM

A conventional activity log can show that an agent called a tool. It may not show who granted authority, which limit applied, whether the exact terms matched or why the action was permitted at that moment.

WHO IT IS FOR

Risk, audit, security, procurement and operations teams evaluating evidence requirements for consequential AI-agent workflows.

CURRENT BOUNDARY

What it can genuinely support today.

CURRENTLY SUPPORTED OR POSITIONED
  • The fictional walkthrough displays the proposed amount, authority ceiling and resulting decision.
  • An authorised outcome is described as bound to exact recorded terms rather than to a reusable general instruction.
  • Recipient verification is a separate step, so the recipient need not rely on the agent's summary alone.
  • Changed terms and replayed evidence are rejected in the public scenario.
  • A controlled pilot would define retention, identifiers, human accountability and downstream audit correlation for the selected workflow.
WHAT THIS PAGE DOES NOT CLAIM
  • The public website does not expose a production ledger, customer audit records or cryptographic protocol specification.
  • Decision evidence does not prove delivery, acceptance, execution, underlying real-world truth or legal effect.
  • Evidence design must be reviewed against the legal, security, privacy and retention requirements of each deployment.