The evidence file

46 artefacts appear across ISO/IEC 42001. 0 of them are demanded by more than one obligation, which makes them the ones worth building first: the same document, properly made, discharges several duties at once. An artefact that is merely present does not discharge anything, so each entry records what makes it sufficient — that sentence is the part an assessor is testing.

What you will be asked for

• A communication record showing reach rather than circulation

  One obligation

  • Clause 5.2 — AI policy

    Which populations received it, by what means, and what proportion of the roles it governs is covered. A link on an intranet page establishes availability and says nothing about whether anyone was reached.

• A competence determination made before people are assigned

  One obligation

  • Clause 7.2 — Competence

    What each AIMS role must be able to do, written as demonstrable actions rather than as technologies somebody should be familiar with. Written after the team was chosen, it records who was available rather than what was needed.

• A context register keeping external and internal issues apart

  One obligation

  • Clause 4.1 — Understanding the organization and its context

    One list per side, every entry with a named owner and a review date. The internal issues an AI function generates — model estate, skills, tooling dependency, compute — are the ones usually folded into a general enterprise register and lost there.

• A derivation trace from the policy to each AI objective

  One obligation

  • Clause 5.2 — AI policy

    One line per objective naming the part of the policy it answers. Objectives tracing instead to an operational plan are evidence that the document is not functioning as the framework this clause asks it to be.

• A legal and regulatory register mapped to named systems

  One obligation

  • Clause 4.1 — Understanding the organization and its context

    The instruments that bind, the prohibited-use provisions among them, and which system each one reaches. A register naming statutes but no systems cannot demonstrate that any particular use was ever checked against a prohibition.

• A reassessment interval with evidence it was applied

  One obligation

  • Clause 7.2 — Competence

    Capability here decays against a moving subject: a matrix set two model generations ago describes work nobody now does. The artefact is the date of the last reassessment, not the policy saying one happens.

• A reassessment log against the interval the organisation set

  One obligation

  • Clause 8.4 — AI system impact assessment

    Dates the work was due, dates it was done, and the reasons for any gap. A log recording only completions cannot demonstrate that the cadence the organisation chose for itself was honoured.

• A repeatability check performed by a second assessor

  One obligation

  • Clause 6.1.2 — AI risk assessment

    One system handed independently to somebody who did not do the original work, with the two evaluations compared and the difference explained. Almost no organisation runs this, and it is the only direct test of whether the criteria decide anything.

• A role assignment register naming individuals rather than forums

  One obligation

  • Clause 5.3 — Roles, responsibilities and authorities

    Each AIMS role, the person holding it, and the authority attaching to it. Where a row names a committee, an assessor asks which member answers for it, and the register should have settled that in advance rather than at interview.

• A supply-relationship map showing where the line cuts

  One obligation

  • Clause 4.3 — Determining the scope of the AI management system

    For each system crossing a corporate boundary — hosted, licensed, or supplied by a group entity — where the split falls and what governs the far side. Third-party systems are where a scope quietly stops without saying that it has.

• An AI system inventory covering inclusions and exclusions alike

  One obligation

  • Clause 4.3 — Determining the scope of the AI management system

    Every system identified, in or out, and for each one left out a reason and the person who signed it off. An inventory listing only what is inside records the result of a decision and destroys the proof that any decision was taken.

• An effectiveness evaluation for every action taken to close a gap

  One obligation

  • Clause 7.2 — Competence

    A later assessment against the criterion the shortfall was found on, dated after the action. A completion certificate closes an administrative loop and answers nothing this clause puts.

• An interested-party register with the 4.2 filter applied

  One obligation

  • Clause 4.3 — Determining the scope of the AI management system

    Parties, what each of them expects, and the recorded decision on which of those expectations the AIMS will address and which it will not. The third column is the one clause 4.2 actually asks for and the one most registers omit.

• Audit reports with findings graded to a stated methodology

  One obligation

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

    A grading scale defined before the work, applied consistently, and visible in the report. Where every finding is minor and no scale is published, the grading is a negotiation outcome rather than an assessment.

• Auditor independence declarations, one per engagement

  One obligation

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

    For each auditor, what they did not audit and why — the policy they drafted, the register they own, the process they run. A blanket declaration of independence at function level cannot address the conflicts that actually arise in a small team.

• Competence criteria for auditing an AIMS, and evidence against them

  One obligation

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

    What an auditor of this management system must be able to do, written before staffing, with per-person evidence of education, training or experience against it. Attendance at an introductory session evidences attendance.

• Completed assessments, one per system and configuration

  One obligation

  • Clause 8.4 — AI system impact assessment

    Each carrying the version of the system it examined, the date, and a named approver. An assessment tied to a product name rather than to a configuration cannot tell a reader which of several retrained models it was actually about.

• Controls adopted beyond the annex, with their reasoning

  One obligation

  • Annex A and Annex B — Reference control objectives and controls · Implementation guidance for AI controls

    What the organisation added because its own analysis called for it, and why the reference set did not cover the ground. An estate with unusual exposure and no additions suggests the annex was treated as the boundary of what could be considered.

• Dated change history of the boundary

  One obligation

  • Clause 4.3 — Determining the scope of the AI management system

    Successive versions with what moved and why, tied to the context change or new requirement that prompted the revision. A boundary revised only at recertification is behaving as a certificate artefact rather than as a live constraint.

• Evidence of review after an external movement, dated

  One obligation

  • Clause 4.1 — Understanding the organization and its context

    A named change outside the organisation, when it was noticed, and what was reopened because of it — including a reasoned decision that nothing needed to change. The trail of the review is what separates a live document from an archived one.

• Evidence that each assignment was communicated to its holder

  One obligation

  • Clause 5.3 — Roles, responsibilities and authorities

    Acknowledgement by the named person, dated. Assigning and communicating are separate limbs, and the failure mode is a manual allocating a duty to somebody who first learns of it when they are asked about it.

• Operating evidence for every control marked applicable

  One obligation

  • Annex A and Annex B — Reference control objectives and controls · Implementation guidance for AI controls

    Something the control produced while running — a record, a log, an approval, a rejected change — rather than the document that describes it. Adoption and operation are different claims and only the second can be sampled.

• Ownership and last-evaluated date per register entry

  One obligation

  • Clause 6.1.2 — AI risk assessment

    A named owner with authority over the treatment, and the date the entry was last worked on rather than the date it was created. Entries with neither are the ones nobody revisits between certification visits.

• Per-person evidence against the stated criteria

  One obligation

  • Clause 7.2 — Competence

    Education, training or experience mapped criterion by criterion, with what the experience demonstrated and who judged it. Attendance at a session evidences that the session took place.

• Reports of AIMS performance to top management, with dates

  One obligation

  • Clause 5.3 — Roles, responsibilities and authorities

    What was reported, by whom, to whom, and when. An agreed frequency that slipped is a smaller problem than a duty discharged inside somebody else’s meeting and never written down anywhere.

• The AI risk register keeping identification, analysis and evaluation apart

  One obligation

  • Clause 6.1.2 — AI risk assessment

    Three recorded steps rather than one column of ratings. Where identification and evaluation are collapsed, a category such as bias or drift gets scored as though it were a risk and the specific failure mode is never named.

• The alignment analysis against other organisational instruments

  One obligation

  • Clause 5.2 — AI policy

    Which policies were compared, what conflicts emerged, and how each was resolved. Where nothing emerged the record should still show what was examined, because an unrecorded comparison and an unperformed one look identical afterwards.

• The applicability review record after a scope or risk movement

  One obligation

  • Annex A and Annex B — Reference control objectives and controls · Implementation guidance for AI controls

    Dated revisits triggered by a boundary change, a new system, or a material shift in the risk picture, with what changed as a result. A statement whose only version is the original was written for a certificate rather than for use.

• The applicability statement, with a decision and a justification per control

  One obligation

  • Annex A and Annex B — Reference control objectives and controls · Implementation guidance for AI controls

    Two fields that cannot be inferred from one another: what was decided, and the reasoning that produced it. A justification restating the control’s own subject matter as a claim of compliance has recorded an intention, not a determination.

• The approval of record for each assessment result

  One obligation

  • Clause 8.4 — AI system impact assessment

    A named individual with the authority to accept the conclusion, and the date they did. Approval by a working group or a distribution list leaves nobody who can be asked afterwards why the result was accepted.

• The conformity assignment, stated discretely

  One obligation

  • Clause 5.3 — Roles, responsibilities and authorities

    Answering for the management system’s conformity to the standard should appear as one line held by one person, not spread across owners of individual systems who each answer only for their own.

• The consistency argument against enterprise risk criteria

  One obligation

  • Clause 6.1.2 — AI risk assessment

    How the AI criteria relate to the thresholds the organisation already uses elsewhere, and where they deliberately differ. An AI scale calibrated in isolation produces gradings the board cannot compare with anything else it reads.

• The dated adoption record of the policy by top management

  One obligation

  • Clause 5.2 — AI policy

    Who approved it, on what date, and in what forum. An assessor is looking for an act of adoption attributable to a person with authority to bind the organisation, not for a document carrying a version number and an author field.

• The documented risk assessment process, with its criteria

  One obligation

  • Clause 6.1.2 — AI risk assessment

    Method, inputs, who performs it, and the criteria it applies — held as one document rather than reconstructed from a completed register. This is the artefact the clause asks for, and it is separate from anything the process has produced.

• The documented scope statement, approved and dated

  One obligation

  • Clause 4.3 — Determining the scope of the AI management system

    Boundaries written as named systems, the interfaces they expose and the environments they run in, with an approver and a date. A statement expressed in departments or office locations cannot answer the first question an assessor asks, which is on which side of the line a given system falls.

• The internal audit programme, with its risk basis stated

  One obligation

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

    Frequency per process and the reasoning behind each choice, so that a reader can see which parts of the AIMS the organisation considers to matter more. A programme with one frequency for everything has recorded a calendar and no judgement.

• The issues register the boundary was derived from

  One obligation

  • Clause 4.3 — Determining the scope of the AI management system

    External and internal issues, dated before the scope rather than after it. Where nothing in the register failed to make its way into the boundary, the register was almost certainly written to justify a conclusion someone had already reached.

• The linkage record from findings to treatment decisions

  One obligation

  • Clause 8.4 — AI system impact assessment

    Each material finding followed to the control, monitoring measure or acceptance record it produced, with the owner. Where a finding leads nowhere, the record should say that it was considered and closed, and on whose authority.

• The organisation’s position on the Annex B guidance, per control

  One obligation

  • Annex A and Annex B — Reference control objectives and controls · Implementation guidance for AI controls

    Because that annex is normative rather than advisory, what was done with its implementation detail is part of what an assessor may work from — including, for each departure, what was done instead and on what reasoning.

• The previous-results input, showing the programme moved

  One obligation

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

    What the last cycle found and what changed in this cycle because of it — a process brought forward, a method altered, coverage widened. Where nothing changed, the file should record that this was considered and why it was right.

• The reassessment trigger definition, approved and dated

  One obligation

  • Clause 8.4 — AI system impact assessment

    Which events oblige the work to be redone, written before they happen and specific enough to argue about — retraining, threshold change, a new user population, a new jurisdiction. An undefined trigger can never be shown to have fired.

• The record of reporting to relevant management, dated

  One obligation

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

    Who received the results, when, and in what form, including where the recipient was not the person who commissioned the work. Results circulated only within the function audited have not left the boundary they were meant to cross.

• The role determination, per system, with reasoning

  One obligation

  • Clause 4.1 — Understanding the organization and its context

    For each system in the estate, every role held toward it and how that conclusion was reached. A single company-level answer cannot be right where a firm both builds and buys, and the clause says plainly that more than one role may apply.

• The societal-impact section, actually completed

  One obligation

  • Clause 8.4 — AI system impact assessment

    What the system does to people who are neither its users nor its subjects, and to the market or public good it operates inside. This is the section that separates the clause from a privacy exercise and the section most often left blank.

• The trace from treatment decisions to selected controls

  One obligation

  • Annex A and Annex B — Reference control objectives and controls · Implementation guidance for AI controls

    Each control tied back to the risk it treats and the decision that chose it, so selection is a conclusion rather than an inheritance. This is the thread that is missing whenever the statement was filled in before the risk work was finished.

• Threshold statements for each measurable AI risk

  One obligation

  • Clause 6.1.2 — AI risk assessment

    A figure, the population it is measured over, the observation window, and the level at which the result stops being acceptable. Without the figure a grading is a preference, and two people applying it will not land in the same place.