Framework overview
AIE-12 · Autonomy & Impact Evidencev0.1 · Draft for public comment · 2 October 2026

A claim is only as strong as the evidence behind it.

AIT-12 says which controls a deployment needs. AIE-12 says what counts as proof that they are in place: how strong the evidence must be, who must produce it, and how long it stays valid. The higher the horizon, the less an implementer’s word is enough on its own.

AIH-12 classifies. AIT-12 sets the gate. AIE-12 proves.

Six classes

Grading the evidence

1Assertion

A statement by the implementer. Nothing has been checked.

2Demonstration

Shown working, under conditions chosen by the implementer.

3Documented case

A recorded real-world instance, with enough detail for someone else to examine.

4Reproducible test

A test someone else can re-run and get the same result.

5Controlled pilot

Assessed in a bounded live deployment, with stated conditions and published failures.

6Independent review

Examined by a party with no stake in the result, who publishes the findings.

Weakest to strongest. These classes replace the earlier E0–E5 proposal (register item OCC-E-003).

E1 – E12

Twelve evidence profiles

Each profile matches one AIT-12 threshold. To claim Tn, every control in it must be supported to at least profile En.

ProfileSupportsMinimum evidenceIndependenceValid for
E1T1AssertionSelfCurrent at time of claim
E2T2Assertion, plus a Demonstration of disclosureSelf24 months
E3T3DemonstrationSelf24 months
E4T4Reproducible test (core tasks)Self, re-runnable by others12 months
E5T5Reproducible test of interruption and revocation; Documented case for authority limitsSelf, re-runnable by others12 months
E6T6Controlled pilot of chain revocationSecond party recommended6 months
E7T7Independent review of capability evaluationIndependent6 months
E8T8Independent review plus Controlled pilotIndependent6 months
E9T9Continuous independent review · speculativeIndependent and publicContinuous
E10T10Institutional review record · speculativePublic institutionContinuous
E11T11Multilateral review record · speculativeMulti-jurisdictionContinuous
E12T12Not defined · speculativeNot applicableNot applicable
Rules

How evidence is weighed

Every control, its own evidence

A profile is met only when every control in the matching threshold is supported at or above the minimum class. One strong test does not cover a weak one elsewhere.

Label every source

Each piece of evidence carries its class, who produced it, and the date. Unlabelled evidence counts as an assertion.

Evidence expires

Evidence older than the freshness period no longer supports the claim. A material change to the deployment resets the clock.

Weaker evidence is disclosed

Where a control is supported below the minimum class, the claim says so. It is recorded as “evidence below profile”, not hidden.

Failures count as evidence

Published failures from pilots and incidents are part of the record. A profile that only ever shows passes is incomplete.

Uncertainty is stated

Every evidence statement says what could not be checked. Unknowns are recorded as unknowns.

Why this name
AIE-12 completes the family on the same pattern as its siblings: Autonomy & Impact Horizons, Thresholds, Evidence. One horizon, one threshold, one evidence profile — Hn, Tn, En. The attestation format for recording this evidence is the planned Evidence & Attestation profile.