Verified Measurement Credential: What It Means and How to Verify
A verified measurement credential is a signed, auditable assertion that a specific measurement was taken at a specific time by a specific instrument, and that the result has not been altered since. The Council of AI issues verified measurement credentials for AI-system evaluations across 14 axes: governance, privacy, AGI-readiness, ASI-readiness, MCP interoperability, open-source compliance, machine reasoning, care/safety, extended reality, detection/interop, Article 5 screening, swarm coordination, and affect alignment. Each credential carries an Ed25519 signature over a canonical body hash, with a published public key for independent verification.
To verify a credential: (1) obtain the signed JSON artifact, (2) run the Council of AI sign.py verifier (sign.py --verify <path>), (3) confirm the output reads VALID and was signed by the published key, (4) check the body_sha256 matches the canonical body you received. This is not a security guarantee -- it is a measurement guarantee. The credential says: this board was measured, by this fleet, at this time, and the numbers are what they are.
It does not say the model is safe -- it says the safety was measured and the measurement is unaltered. For regulators and procurers, this audits the measurement pipeline without requiring trust in the measuring lab. For lab operators, it provides a tamper-evident record of every evaluation run.
The credential format is designed for PQC upgrade (ML-DSA-65 via liboqs) using the same body and a swapped primitive, ensuring forward compatibility.
Standards and sources referenced
- Council of AI sign.py (open-source verifier)
- DEFONEOS-SEAL standard (csoai-defoneos.org)
- CSOAI Verified Measurement Credential specification
- NIST IR 8475 (digital credential formats)
- ISO/IEC 42001 AI Management System audit requirements