The original title is: "Agent Evidence Levels (AEL): grading the evidence your AI agent leaves behind"

Short summary

Agent Evidence Levels (AEL) is an open standard for grading how much of an AI agent's behavior can be verified by outside parties without trusting the vendor. It spans AEL-0 (signed records) to AEL-4 (counterparty confirmation), with each level adding independent verification mechanisms to catch omissions. The framework includes a reference checker and open-source conformance corpus.

• •AEL grades agent auditability across five levels, from basic signed and hash-linked records to cross-domain, externally anchored, and counterparty-confirmed evidence
• •Each level adds independent verification mechanisms (signatures, gap detection, external append-only logs, destination attestation) to detect omissions without trusting the operator
• •Open-source reference implementation ships with the standard; author designed it for eventual donation to neutral governance once vocabulary matures

Generated with AI, which can make mistakes.