Aegentra Labs

Free, no account required

ISO/IEC 42001 Lead Implementer practice questions

Question 1 tests how controls are determined against Annex A. Questions 2–5 follow a live-captioning company through its AI system impact assessment, a nonconformity, data provenance for a new model and evidence that human oversight works. Reveal each answer to see the reasoning and why the other options fail.

These practice questions are independently authored by Harry Sidhu and Aegentra and are not official PECB examination materials. They do not reproduce confidential or recalled PECB examination content. Read the methodology.

Questions
5
Options
3 each
Answers
Explained
Start question 1

What this exact sample covers

AIMS implementation decisions

  • Annex A and the Statement of Applicability
  • Impact assessment and data provenance
  • Corrective action and oversight evidence
  1. Question 01Domain 3 · analysis

    An online travel agency determined eleven controls for its AI risk treatment from its corporate control library. Its Statement of Applicability lists those eleven controls with their implementation status and states that Annex A of ISO/IEC 42001:2023 was not used because the company prefers its own controls. Is this in accordance with clause 6.1.3?

    1. ANo, because clause 6.1.3 requires the determined controls to be compared with Annex A and the SoA to justify the inclusion or exclusion of controls.
    2. BYes, because clause 6.1.3 allows controls from any source, and Annex A is a list of examples that the organization is free to disregard entirely.
    3. CNo, because clause 6.1.3 requires every Annex A control to be implemented, so the library controls can only supplement the full Annex A set.
    the answer and reasoning

    Correct answerA. No, because clause 6.1.3 requires the determined controls to be compared with Annex A and the SoA to justify the inclusion or exclusion of controls.

    Clause 6.1.3 lets the organization design or source its own controls, but it shall compare them with Annex A to verify that no necessary controls have been omitted and produce a Statement of Applicability that justifies the inclusion and exclusion of controls. An SoA that ignores Annex A fails both steps. Annex A is normative, yet its controls need not all be implemented.

    Why the other options fail

    • B. Annex A is a normative reference set in ISO/IEC 42001:2023; clause 6.1.3 requires the comparison with it and justified exclusions, so it cannot simply be disregarded.
    • C. Clause 6.1.3 does not require every Annex A control to be implemented; controls may be excluded when the SoA justifies the exclusion, for example because the risk assessment does not call for them.

    Published exam-scope sources: ISO/IEC 42001:2023 — Information technology, Artificial intelligence, Management system · ISO/IEC 42001 Lead Implementer — training course page

  2. Scenario for questions 2–5

    Kerrowen Captioning: live captions and a new model

    Kerrowen Captioning employs 210 people and supplies live and pre-recorded captions to five regional television broadcasters. It develops its own speech-recognition model, which drafts live captions with a three-second delay while trained caption editors monitor the output. Kerrowen is implementing an artificial intelligence management system (AIMS) based on ISO/IEC 42001:2023 that covers both services, with a certification audit planned for next year.

    The AI system impact assessment for live captioning, completed in March, considers deaf and hard-of-hearing viewers, the broadcasting rules of the two jurisdictions where Kerrowen's customers operate, and the effect of caption errors during news and sport. Support tickets since January show that two broadcasters publish the unedited live captions on their websites as searchable programme transcripts, including transcripts of court reporting. The assessment does not mention this.

    The live-captioning procedure requires an editor to take over manual captioning whenever the model's confidence score stays below 0.80 for more than ten seconds; the system logs confidence scores and every takeover with the time, programme and editor. The procedure also requires every emergency broadcast to be captioned manually by a named editor. During a flood warning last month no editor was rostered, and the model misnamed an evacuation route for four minutes. The operations manager corrected the caption on air, reminded the rostering team of the rule and closed the record.

    The data team plans to release model version 5 in six weeks. It was retrained on 2,600 hours of audio, including 700 hours bought from a broker whose invoice describes the material as ‘licensed broadcast speech’; the broker supplied no record of where or how the recordings were obtained. Version 5 reduced the word error rate on Kerrowen's validation set from 7.1% to 5.4%.

    Question 02Domain 3 · analysis

    Based on the scenario, which requirement of clause 6.1.4 of ISO/IEC 42001:2023 does Kerrowen's AI system impact assessment for live captioning not yet meet?

    1. ADetermining the consequences of foreseeable misuse, since the publication of unedited captions as searchable transcripts is known but has not been assessed.
    2. BTaking applicable jurisdictions into account, since the assessment must cover every country in which a published website transcript could be read.
    3. CConsidering consequences for individuals, since the assessment looks at groups of viewers as a whole rather than at each individual viewer.
    the answer and reasoning

    Correct answerA. Determining the consequences of foreseeable misuse, since the publication of unedited captions as searchable transcripts is known but has not been assessed.

    Clause 6.1.4 requires the AI system impact assessment to determine the potential consequences of the system's deployment, intended use and foreseeable misuse for individuals, groups and societies. Kerrowen knows from support tickets that raw captions are republished as transcripts, including court reporting, yet the assessment omits that use. Jurisdictions of deployment and affected viewers are already covered.

    Why the other options fail

    • B. The assessment already considers the two jurisdictions where the system is deployed; clause 6.1.4 does not extend this to every country where a transcript might be read.
    • C. Clause 6.1.4 refers to individuals or groups of individuals, or both, so assessing deaf and hard-of-hearing viewers as a group already meets this element.

    Published exam-scope sources: ISO/IEC 42001:2023 — Information technology, Artificial intelligence, Management system · ISO/IEC 42001 Lead Implementer — training course page

  3. Question 03Domain 6 · evaluation

    The operations manager corrected the flood-warning caption on air, reminded the rostering team of the rule and closed the record. Is this in accordance with clause 10.2 of ISO/IEC 42001:2023?

    1. AYes, because the caption was corrected within minutes and the reminder deals with the rostering lapse, which together make up the corrective action.
    2. BNo, because clause 10.2 requires the nonconformity to be reported to the broadcasters and to a regulator before its record can be closed.
    3. CNo, because clause 10.2 also requires evaluating action to remove the cause, checking for similar cases and reviewing the effectiveness of action.
    the answer and reasoning

    Correct answerC. No, because clause 10.2 also requires evaluating action to remove the cause, checking for similar cases and reviewing the effectiveness of action.

    Captioning an emergency broadcast without a named editor in manual mode breached Kerrowen's own procedure, so it is a nonconformity. Clause 10.2 requires reaction and correction, then evaluation of the need for action to eliminate its causes, a check for similar nonconformities, implementation of any action needed and review of its effectiveness. A correction and a reminder stop short of that.

    Why the other options fail

    • A. The on-air fix is a correction and the reminder does not establish why no editor was rostered; clause 10.2 requires the cause to be evaluated and the effectiveness of action reviewed.
    • B. Clause 10.2 contains no requirement to notify broadcasters or a regulator before closure; any such duty would come from contracts or law rather than from this clause.

    Published exam-scope sources: ISO/IEC 42001:2023 — Information technology, Artificial intelligence, Management system · ISO/IEC 42001 Lead Implementer — training course page

  4. Question 04Domain 4 · evaluation

    Before releasing model version 5, what should Kerrowen do first about the 700 hours of audio bought from the broker?

    1. ARelease version 5 as planned, because the lower word error rate on the validation set shows that the brokered audio is of suitable quality.
    2. BObtain and record the provenance of the brokered audio, including its source and collection basis, and then decide whether it may be used.
    3. CRely on the description on the broker's invoice, because buying licensed material transfers responsibility for its origin to the broker.
    the answer and reasoning

    Correct answerB. Obtain and record the provenance of the brokered audio, including its source and collection basis, and then decide whether it may be used.

    Control A.7.5 in Annex A of ISO/IEC 42001:2023 requires a process for recording the provenance of data used in AI systems, and A.7.3 covers how data is acquired. Kerrowen cannot show where 700 hours of training audio came from or how it was collected, so provenance must be established before deciding to use it. Accuracy gains and purchase terms do not answer that question.

    Why the other options fail

    • A. Word error rate measures model performance, not where the recordings came from or whether Kerrowen may lawfully and responsibly use them in training.
    • C. Buying data does not transfer Kerrowen's accountability for the data used in its own AI system, and an invoice line is not a provenance record.

    Published exam-scope sources: ISO/IEC 42001:2023 — Information technology, Artificial intelligence, Management system · ISO/IEC 42001 Lead Implementer — training course page

  5. Question 05Domain 5 · application

    Which evidence would best demonstrate that Kerrowen's editor takeover measure for live captioning operates as designed?

    1. AThe approved live-captioning procedure that sets the 0.80 confidence threshold and the ten-second rule for editor takeover of live captions.
    2. BTakeover logs matched to the confidence records for sampled programmes, showing each sustained drop below 0.80 followed by a takeover.
    3. CTraining records showing that every caption editor completed the takeover module and passed its assessment within the last year.
    the answer and reasoning

    Correct answerB. Takeover logs matched to the confidence records for sampled programmes, showing each sustained drop below 0.80 followed by a takeover.

    Evidence that a human oversight measure operates must show it working in practice. Matching logged confidence drops against logged takeovers for a sample of programmes tests whether every triggering condition led to the required intervention, which supports monitoring under clause 9.1. A procedure shows design and training records show competence; neither shows the measure operated.

    Why the other options fail

    • A. The procedure documents how the measure is designed to work; it gives no evidence that editors actually took over when the threshold was breached.
    • C. Training records show editors were prepared to intervene, not that takeovers happened each time the confidence score fell below the threshold.

    Published exam-scope sources: ISO/IEC 42001:2023 — Information technology, Artificial intelligence, Management system · ISO/IEC 42001 Lead Implementer — training course page

Preparing for a different exam?

Every live exam has its own five free questions, written for that exam's domains.