Advanced · Safety

Agent Action Evidence

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. Verifiable records proving what an agent did, under whose authority, with what inputs and result.

Technical Definition

Verifiable records proving what an agent did, under whose authority, with what inputs and result.

How it works

Action evidence turns agent behaviour from anecdote into fact. Each executed action emits a record containing the agent identity, the delegation chain, the intent that motivated it, the exact parameters, the observed result, and a timestamp — ideally signed and tamper-evident. Evidence supports incident response (what exactly did the compromised agent touch?), dispute resolution in agent-to-agent commerce, and regulatory audit. It is distinct from logging in that it is designed to be verified by a third party.

Related Concepts

• Agent Authorization Protocol (AAP) — A protocol that determines which actions an authenticated agent is permitted to perform, on whose behalf, and within what limits.
• Agent Trust — The calibrated confidence a system or person places in an agent, grounded in verified identity, bounded authority, and observed behaviour.
• Agent Audit Trail — The chronological, tamper-resistant record of an agent's decisions, tool calls, and context sources.
• Agent Transparency — Making an agent's identity, authority, reasoning, and actions legible to the humans and systems it interacts with.