About · Framework v0.2.5 · Assessment 0.2.5-1

The AI Execution Integrity Framework

AIEF is an independent, implementation-agnostic framework published at aief.dev. It defines the evidence and verification properties an AI system must exhibit for its executions to be examinable after the fact. It specifies required properties — never a particular vendor, artifact format, SDK, storage backend or cryptographic architecture.

What AIEF does not assess

AIEF is not concerned with whether an AI output is correct, fair, safe or lawful. It does not evaluate model quality, bias, alignment or regulatory compliance. It assesses one question: can what the system did be evidenced and independently verified?

Four cumulative levels

Conformance is cumulative. A level is only supported when every requirement at that level and all levels beneath it is supported; a failure at a lower level blocks every level above it.

1. 1

   Artifact Capture

   Material executions produce structured records containing enough context to understand what occurred.

   Introduces AIEF-01

2. 2

   Tamper-Evidence + Deterministic Verification

   A defined protected set is integrity-protected at issuance and checked by a repeatable PASS/FAIL verification procedure.

   Introduces AIEF-02, AIEF-03

3. 3

   Portability + Independent Validation

   Evidence can be exported, retained and verified by an independent party without privileged access.

   Introduces AIEF-05, AIEF-08

4. 4

   Agent Chain + Dependency Traceability

   Multi-step chains and material external dependencies are recorded and their integrity is verifiable.

   Introduces AIEF-06, AIEF-07

AIEF-04, AIEF-09, AIEF-10 are assessed but do not gate levels. AIEF-10 is an optional higher-assurance extension.

The ten controls

How scoring works

Scoring is fully deterministic. A rules engine evaluates your answers: a control is Satisfied only when every critical requirement is answered Yes; a critical No makes it Not satisfied; partial implementation makes it Partially satisfied; unanswered or unsure critical requirements return Insufficient information. Controls AIEF-06 and AIEF-07 may be Not applicable when the system genuinely has no external dependencies or is single-step. Only these rules determine control statuses and levels.

AI is used solely to explain the result in plain language. It cannot change a control status, an applicability decision, the blocking control list or your level.

Self-reported versus demonstrated

This tool reports self-reported readiness. Claiming a capability is not the same as demonstrating it: demonstrated conformance requires examination of actual artifacts, verifier behaviour and independent verification output. Your report lists the evidence you would need for each control.

AIEF and other frameworks

AIEF is focused specifically on AI execution integrity: whether what an AI or agent actually did can be reconstructed, attributed, preserved and independently verified. Other frameworks address broader concerns — governance, risk, safety, management systems, regulatory obligations, privacy, fairness and organisational controls.

AIEF complements those frameworks rather than competing with them. It defines how execution evidence can be captured, protected and independently verified, so that evidence can feed into a broader regulatory, assurance or management-system programme. External frameworks map to AIEF controls; they never change what AIEF conformance means, and AIEF readiness never establishes compliance with an external regime.

View framework crosswalks

This framework and assessment tool are provided for informational purposes only. Results are indicative, based on self-reported information, and do not constitute independent verification, certification, attestation or legal advice.