AIAC-09 · Advanced endorsement
For practitioners assuring systems that take actions rather than produce outputs. Authorisation boundaries, reversibility, and audit trails for software that acts.
§ 1 — Intended candidates
Experienced assurance, risk, and engineering practitioners responsible for systems that execute actions, hold credentials, or spend money.
§ 2 — Examination domains
Domain weightings are published with the full syllabus ahead of the first cohort.
Defining what an agent may do, with whose authority, and how that boundary is enforced rather than requested.
Classifying actions by whether they can be undone, and constraining those that cannot.
Why an agent cannot be unit-tested, and what evaluation of a system that fails differently each run looks like.
Reconstructing what an agent did, why, and on whose instruction — including across tool calls.
Placing a human in the loop where it changes the outcome, rather than where it is easiest to add.
§ 3 — How to certify
Preparation is optional. Choose whichever route suits how you already work — the examination, the fee, and the marking standard are the same either way.
Register and sit the examination. Suitable for practitioners already working in assurance who want the credential without preparation.
The same examination, bundled with self-paced preparation mapped to the syllabus. Taking it confers no advantage in marking and is never required.
Preparation is optional. The examination is available at the same fee and marked to the same standard whether or not preparation was purchased, and assessors are not told which route a candidate took. How we manage this conflict.
§ 4 — Assessment
Scenario-based examination requiring an oversight design for a supplied agentic system: authorisation boundaries, and which actions are gated because they cannot be undone.
Successful candidates receive a credential with a unique identifier that any employer can check against the public record. See credential verification.
Curriculum mapping indicates alignment of learning content with published frameworks. It does not constitute endorsement by, or certification under, any standards organization or regulator.
§ 5 — Further reading
Producing credible, independently reviewable evidence that an AI system works as intended — and how that differs from governance and compliance.
Three terms used interchangeably that mean different things, have different outputs, and require different independence. A practitioner-level distinction.