Open · Public · Normative

The certification standard for human oversight of AI.

COHESION publishes a single, open specification for how human oversight of high-risk AI systems should be measured, maintained, and audited. The full normative document is below.

Adjacent frameworks (EU AI Act Article 14, NIST AI RMF, Colorado SB 26-189, OMB M-25-21) impose oversight obligations but specify no measurement methodology. This document defines one.

Three certification tiers

L1

L1 Monitoring

Continuous measurement of human oversight behaviour against the seven JIS dimensions. Audit log, decay-curve tracking, and per-decision telemetry.

For: Teams beginning to instrument oversight.

L2

L2 Maintenance

L1 plus the Invisible Maintenance Protocol -- calibrated stimuli embedded in workflow that counteract measured decay before it ships into a regulator-visible decision.

For: High-risk deployments under active regulatory scrutiny.

L3

L3 Certified

L1 + L2 plus third-party conformity assessment. The only tier eligible to apply the COHESION Certified mark. Reviewed against the full normative specification.

For: EU AI Act Article 14, Colorado SB 26-189, OMB M-25-21 regulated operators.

Version
1.2
Status
Open · Public · Normative
Published
2026-04-16
Last revised
2026-04-28
Custodian
COHESION AUTH LLC (Washington, USA)
Licence
Permissive open-standard licence. Implementation is free. The COHESION Certified mark is trademarked.

Specification outline

Plus 14 annexes (A through N) and supporting reference sections. Full normative text at /standard/spec/.