How does Council Signal measure AI governance?

Every governance assertion is reduced to a measurable item, scored by a multi-model board across 14 independent axes to produce a spread rather than a single number, and signed with an Ed25519 credential any third party can verify without access to the scoring models.

Council Signal: How Governance Measurement Works

Council Signal is the measurement backbone of the Council of AI. It answers one question: how do you verify that an AI governance claim is true, not just asserted? The protocol is simple in principle: every governance assertion (compliance, safety, containment, provenance) is reduced to a measurable item, scored by a multi-model board, and signed with an Ed25519 credential that any third party can verify without access to the scoring models. The board runs 14 independent axes -- from detector-interop to cross-risk, from Article 5 prohibited-practice classification to jail resilience -- each measured across multiple models to produce a spread, not a point estimate.

A governance score is not a single number but a distribution: the spread between the strongest and weakest model on an axis IS the measurement. Narrow spread means consensus; wide spread means uncertainty. The verified measurement credential (DEFONEOS-SEAL) attests that the measurement was performed by a registered board, that the methodology was documented, and that the result is reproducible.

No council vote, no opinion, no self-assessment: measured, signed, verifiable.

Standards and sources referenced

Council of AI board measurement methodology (14-axis multi-model board, August 2026)
CSOAI Verified Measurement Credential standard (DEFONEOS-SEAL)
ETSI EN 304 223: Cybersecurity for AI (13 principles, January 2026)