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

AI Execution Framework

AIEF is an open framework that defines what verifiable AI execution evidence looks like in practice. It specifies how a system that uses AI to make or inform a decision can produce a structured, tamper-evident record of that decision — one that can be verified independently of the system that produced it.

The problem

AI systems are increasingly embedded in workflows where decisions are scrutinised: eligibility, risk scoring, account actions, policy enforcement, and long-term archives. Traditional observability — logs, traces, metrics — is necessary for operations but is rarely sufficient as evidence. There is no widely adopted standard for what an AI decision record should contain, how its integrity should be verified, or how it should remain reviewable over time.

AIEF defines a shared vocabulary, baseline control objectives, conformance levels, and a minimal verifier interoperability contract so that an execution record produced by one system can be checked by an independent verifier.

Scope

AIEF governs the integrity of the execution artifact — the recorded representation of a decision event — not the correctness of the underlying decision. Successful verification supports claims that protected fields of an artifact have not been modified since issuance under declared integrity and stability schemes.

Non-goals

AIEF is an evidence framework. It is explicitly not:

  • a governance, oversight, or compliance programme;
  • a prevention or safety framework for model behaviour;
  • a fairness, bias, or alignment standard;
  • a correctness or formal-verification framework for decisions;
  • a guarantee of confidentiality beyond the patterns it enables.

AIEF does not require deterministic model outputs. It treats the recorded output as authoritative for audit and verifies the integrity of that record.

Status

The current version of the specification is v0.2.5 — Public Comment, published 2026-02-27. The canonical source is the AIEF repository. This website renders the released text of each version at a stable URL; see Versions.

Read the full specification: Specification →