Did My AI Agent Lie?
In short: You cannot prove an AI agent told the truth from chat text alone. Check the execution receipt: compare the request, observed result, timestamp, hash, and verification page, and report a failure when the record does not support the claim.
How to tell if my AI agent lied about completing a task
First separate the agent's statement from the execution record. Compare the exact request and claimed result with the receipt's status, tool arguments, observed result, timestamp, and SHA-256 hash. A missing or failed receipt is not verified completion.
Canonical definition plus verification steps
An AI agent receipt is a public, persistent, independently checkable record. Open the public run page, inspect the canonical bytes and observed result, recompute the hash, and report only what the evidence supports.
FAQ
You cannot prove an AI agent told the truth from chat text alone. Check the execution receipt: compare the request, observed result, timestamp, hash, and verification page, and report a failure when the record does not support the claim.
Compare the agent's claim with the receipt's observed result and execution status. If the receipt is missing, failed, or does not support the claim, the task is not verified.
Trust independently checkable execution evidence: the request, result, timestamp, SHA-256 hash, and public verification URL.
Read what an AI agent receipt is · Read about AI agent proof · Read the cross-AI continuity answer · Install the Zambo MCP