Where this goes wrong

A weak answer is rarely a wrong one. It is usually a true statement doing less work than it appears to — a policy that exists but binds nobody, an artefact produced once and never since, a judgement recorded without the person who made it or the basis they made it on. Each entry below names the specific failure for its provision; the obligation page it links to sets out what discharges the duty instead.

The answer that does not hold

• Annex A and Annex BReference control objectives and controls · Implementation guidance for AI controls

  The applicability statement inherited from a consultant. Every control appears, every justification is the control’s own subject matter turned into a sentence of compliance, and the exclusions column is empty because excluding nothing felt safer. It reads as complete and evidences nothing: there is no route from any assessed risk to anything on it, so the document records what the standard lists rather than what the organisation decided. The version history usually settles the matter — a single entry, dated the week the certification programme began.

• Clause 4.1Understanding the organization and its context

  The enterprise context register with "AI" appended. Somebody takes the SWOT already used for another management system, adds two bullets about generative models and a skills shortage, and files it as the 4.1 record. Nothing in it determines a role, nothing maps a legal requirement to a system, and nothing in it would read differently if the organisation stopped using AI tomorrow. The tell is the absence of any entry that could turn out to be wrong: a register no external event could falsify is a description of an industry, not a determination about a company.

• Clause 4.3Determining the scope of the AI management system

  The universal scope, adopted to avoid the awkwardness of leaving anything out. "All AI activities of the group" excludes nothing, therefore derives from nothing, and can be tested by nobody, because there is no case it decides. It also commits the organisation to every system anyone procures next quarter, including the ones it has not yet heard about. Between that and a departmental boundary there is a line somebody actually reasoned about, and it is the reasoning rather than the width that clause 4.3 asks to see.

• Clause 5.2AI policy

  The charter of principles offered as the policy. A page of commitments — fairness, accountability, transparency, human oversight — endorsed at board level, published, and genuinely believed. It fails on one question, and always the same one: name an AI objective it produced. Nothing in it can be exceeded, missed or scoped, so it binds the organisation to a disposition instead of to a framework, and the objectives that do get set turn out to have been written by whoever owned the budget. The document is not wrong. It is doing a different job.

• Clause 5.3Roles, responsibilities and authorities

  An organogram offered as an assignment. It is accurate, it is current, and it answers a different question: who somebody works for, rather than what they answer for and what they may decide. Nothing on it names conformity with the standard, nothing names whoever reports performance upward, and no box carries an authority — so the one thing the clause exists to establish, that a named person is able to act, is exactly what cannot be read off it. It is usually produced within a minute of the request, which is what makes it convincing.

• Clause 6.1.2AI risk assessment

  The security register with the word "model" inserted. Confidentiality, integrity and availability carry over, a handful of AI-flavoured entries go in at the bottom, and the organisation reports that AI risk is assessed. It is not laziness — the borrowed methodology is mature, already audited, and understood by the committee that will read the output. What it cannot express is a system degrading quietly on one subpopulation while every surrounding control operates correctly, because the frame has nowhere to put a harm with nobody behind it.

• Clause 7.2Competence

  The completion rate reported as competence. Ninety-four per cent of staff finished the AI module, the figure reaches the governance forum every quarter, and no document anywhere states what any role in the management system was required to be able to do. The tell is that the same slide would work unchanged for an information-security module or a fire-safety one, because it measures a training programme rather than the work. This clause asks about the work first and the people second.

• Clause 8.4AI system impact assessment

  The DPIA with a fresh cover sheet. A data protection impact assessment is reused, an "AI" heading is added, and the organisation reports impact assessment as covered. It is usually a competent document answering a narrower question — lawful basis, necessity, proportionality, the rights of a data subject — and it stops at the boundary of the individual. Nothing in it reaches groups as groups and nothing reaches society. The reuse also tends to be one-off, because the instrument it was borrowed from is a pre-processing exercise, so the recurrence this clause is built around never starts at all.

• Clause 9.2, with 9.2.1 and 9.2.2Internal audit · General · Internal audit programme

  The certification rehearsal offered as an internal audit. It is scheduled to land just before the external visit, it walks the clauses in numerical order, and everything raised is drafted so it can be closed before the assessor arrives. That is useful preparation and it is not what the clause asks for, because it tests readiness rather than conformity and structurally cannot produce an inconvenient result. The signature is a report containing nothing anybody had to argue about, in a management system that has not yet been operated for a full cycle without an audience.