AIEF
AI Execution Framework
An open framework for verifiable AI execution evidence.
Version 0.2.5 · Public Comment
Conformance

Demonstrating conformance

AIEF defines conformance in terms of the execution artifact and its verification. An implementation is considered conformant at a given level if its execution artifacts satisfy the control objectives for that level and pass the deterministic verification procedure for the declared integrity and stability schemes.

Conformance levels

The specification defines incremental conformance levels so that implementations can adopt AIEF progressively. The practical starting point is Level 2 (Artifact + Tamper-Evidence + Deterministic Verification). For agentic or multi-step workflows, Level 2 plus chain and dependency traceability (Level 4) is the practical starting point.

The normative definitions of each level, the protected set, the stability scheme, and the verifier interoperability contract are in the Specification → Conformance Levels.

Verification rules

Verification is deterministic. A verifier accepts an execution artifact, the declared stability and integrity schemes, and any referenced evidence material, and returns PASS or FAIL together with a reason code. Reproducing probabilistic model outputs is not part of verification; AIEF uses cache-as-truth semantics for recorded outputs.

Reason codes

The specification enumerates the reason codes that a conformant verifier returns, covering missing fields, broken integrity over the protected set, unknown or unsupported schemes, broken chain links, and other defined failure modes. See Specification → Reason Codes.

Test vectors

Published conformance test vectors accompany the specification in the canonical repository. An implementation that follows the rules in the specification and passes the published test vectors for its declared conformance level is considered conformant for that level.

Test vectors are released alongside each version of the specification and are tagged together; see Versions.