AIGP domain - 25% of the exam

Understanding how laws, standards and frameworks apply to AI

Understanding how laws, standards and frameworks apply to AI is 25% of the AIGP exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

79
Practice questions
6
Objectives
38
Medium
41
Hard

Where people go wrong

  • Applying high-risk duties to lower tiers

    The wrong answers carry conformity assessment, registration in the EU database and notified-body audits across to chatbots, deepfakes and back-office tools. Those duties belong to the high-risk tier. Limited-risk systems owe transparency: people are told they are dealing with AI, and synthetic or manipulated content is marked, with a tailored exception for evidently artistic or satirical works. Minimal-risk systems such as an inventory optimiser face no mandatory duties under the Act, only encouragement to follow voluntary codes.

    See it in a sample question
  • Misplacing the Act's graded fine ceilings

    The wrong answers reach for the GDPR figures of 20 million euros or 4 per cent, put an information offence in the top band, or describe one flat amount. The Act grades its ceilings. Breaching the Article 5 prohibited practices carries up to 35 million euros or 7 per cent of worldwide annual turnover. Other obligation breaches carry up to 15 million euros or 3 per cent. Supplying incorrect, incomplete or misleading information carries up to 7.5 million euros or 1 per cent. Each ceiling is the higher of the fixed sum or the percentage.

    See it in a sample question
  • Leaving provider duties with the original supplier

    The wrong answers keep responsibility with whoever built the component, or treat a CE marking or the earlier assessment as still covering a changed system. A start-up that rebrands a high-risk component, and a firm or hospital that substantially modifies a system or changes its intended purpose, becomes the provider. It must then draw up technical documentation before market placement and carry out a fresh conformity assessment before the changed system is put into service.

    See it in a sample question
  • Confusing the ISO/IEC AI standards

    The wrong answers pick a neighbouring number because each ISO/IEC AI standard sounds relevant to governance. ISO/IEC 42001 specifies requirements for an AI management system that an organisation can audit and certify against. ISO/IEC 42005 gives guidance on AI system impact assessment. ISO/IEC 22989 supplies AI concepts and terminology. ISO/IEC 23894 offers AI risk guidance rather than a certifiable management system, and ISO/IEC 23053 frames how machine learning systems are engineered.

    See it in a sample question
  • Deciding controller status from hosting or contract

    The wrong answers settle the GDPR role from visible features: who hosts the servers, who signed the data processing agreement, or who faces data subjects. Controller status turns on who determines the purposes and means of processing, and a vendor acting on documented instructions is a processor. The split carries into breaches: the processor notifies the controller without undue delay, and the controller assesses the risk, notifies the supervisory authority and, where the risk is high, tells the affected individuals.

    See it in a sample question

Objectives in this domain

What to study

The official documents this domain's practice questions cite most.

Sample questions from this domain

Free sampleUnderstanding how laws, standards and frameworks apply to AIhard

An organisation deploys a customer-facing chatbot and also publishes marketing videos containing realistic AI-generated footage of a fictional spokesperson. Its counsel asks what the EU AI Act's transparency rules require for systems that are not classified as high-risk. Which obligation applies here?

  • APeople must be informed that they are interacting with an AI system, and artificially generated or manipulated audiovisual content must be marked as such. Correct
  • BThe chatbot and the videos must each undergo a full conformity assessment and be registered before any public release in the EU.
  • CNo disclosure is required as long as the chatbot and the synthetic videos are accurate and not misleading to the audience.
  • DThe organisation must appoint an external auditor to approve each chatbot response and every video clip before it is shown to the public.
Identify the EU AI Act transparency duties: disclose interaction with AI and label artificially generated or manipulated content. Beyond the high-risk regime, the EU AI Act sets transparency obligations for certain systems regardless of risk classification. Providers and deployers must ensure that people are informed when they interact with an AI system, unless that is obvious from the circumstances, and that artificially generated or manipulated image, audio, or video content, including deepfakes, is disclosed as such. These duties are about informing the people affected; they are not conformity assessments, are not waived by the content being accurate, and do not call for an external auditor to clear each output.

Why A is correct: The Act's transparency duties require that individuals are told when they are dealing with an AI system unless it is obvious, and that synthetic or manipulated media is disclosed as artificially generated, which fits both the chatbot and the videos.

Why B is wrong: Conformity assessment and registration are tempting because they sound like the rigorous default, but they belong to the high-risk regime, whereas these systems fall under the lighter transparency obligations instead.

Why C is wrong: An accuracy-based exemption is appealing, but the transparency duty exists precisely so people know they are dealing with AI and synthetic media even when the content is accurate, so accuracy does not remove the obligation.

Why D is wrong: External pre-approval sounds like a strong safeguard, but the Act imposes no such per-output auditor for these systems; it requires disclosure to the people affected, not external sign-off on every item.

Free sampleUnderstanding how laws, standards and frameworks apply to AIhard

A large company is found to have deployed an AI system that uses subliminal techniques to materially distort consumer behaviour, a practice falling under the EU AI Act prohibitions. Its counsel is advising the board on the maximum administrative fine the company faces under the Act's penalty framework. Which ceiling applies to this infringement?

  • AUp to 15 million euros or 3 percent of total worldwide annual turnover for the preceding financial year, whichever is higher.
  • BUp to 7.5 million euros or 1 percent of total worldwide annual turnover for the preceding financial year, whichever is higher.
  • CUp to 35 million euros or 7 percent of total worldwide annual turnover for the preceding financial year, whichever is higher. Correct
  • DUp to 20 million euros or 4 percent of total worldwide annual turnover for the preceding financial year, whichever is higher.
Recall that infringing the EU AI Act prohibited practices carries the top penalty band of up to 35 million euros or 7 percent of worldwide annual turnover. The AI Act sets tiered administrative fines, and the most severe tier attaches to breaches of the Article 5 prohibitions. That ceiling is up to 35 million euros or 7 percent of total worldwide annual turnover for the preceding year, whichever is higher, which sits above the 15 million or 3 percent tier for operator obligations and the 7.5 million or 1 percent tier for misleading information.

Why A is wrong: This tier is genuine and therefore tempting, but it applies to breaches of operator obligations such as high-risk requirements, not to the prohibited practices in Article 5, so it understates the ceiling for a banned practice.

Why B is wrong: This band exists but is reserved for supplying incorrect, incomplete or misleading information to authorities, so it is far too low for a prohibited practice and confuses an information-reporting failure with a substantive ban.

Why C is correct: The AI Act reserves its highest penalty band for infringements of the Article 5 prohibitions, set at up to 35 million euros or 7 percent of worldwide annual turnover, whichever is higher, which is the ceiling for using a banned subliminal-manipulation system.

Why D is wrong: These figures echo the familiar GDPR maximum, which makes them feel right by association, but the AI Act does not use a 20 million or 4 percent band, so importing the GDPR ceiling is a misclassification.

Free sampleUnderstanding how laws, standards and frameworks apply to AIhard

A start-up integrates a third-party high-risk AI component into its own product, rebrands the combined product under its own name, and places it on the EU market. Regulators later request the technical documentation for the high-risk system. Under the EU AI Act, who must have drawn up that documentation, and what is its purpose?

  • AThe original component supplier alone must hold the documentation, and its purpose is to shield the start-up from any further compliance responsibility.
  • BAny deployer of the system must compile the documentation after deployment, and its purpose is to record how individual end users configured the product.
  • CThe provider must draw up technical documentation before the system is placed on the market and keep it current, so authorities can assess the system's conformity. Correct
  • DThe notified body must author the documentation during its review, and its purpose is to replace the provider's own internal quality records.
Recognise that the provider must prepare and maintain technical documentation before market placement, and that rebranding a high-risk system makes the rebrander the provider. For high-risk systems the EU AI Act places the technical documentation duty on the provider, who must draw it up before the system is placed on the market and keep it up to date. The documentation demonstrates that the system meets the requirements and gives competent authorities the information they need to assess conformity. A party that puts a high-risk system on the market under its own name or trademark is treated as the provider and assumes these duties, so a rebranding start-up cannot leave the documentation with the original supplier, push it onto deployers, or expect a notified body to author it.

Why A is wrong: Pointing back to the supplier is tempting because it built the component, but by placing the system on the market under its own name the start-up takes on provider duties, so the documentation responsibility does not rest with the supplier alone.

Why B is wrong: End-user configuration logs are a reasonable-sounding artefact, but the technical documentation is a provider duty completed before market placement, not a post-deployment record assembled by deployers.

Why C is correct: The Act requires the provider to prepare technical documentation prior to market placement and maintain it, giving regulators the means to verify that the high-risk system meets the requirements; rebranding makes the start-up that provider.

Why D is wrong: It is tempting to assign authorship to a notified body because such bodies examine documentation, but they assess what the provider has drawn up rather than author it, and they do not replace the provider's records.

Free sampleUnderstanding how laws, standards and frameworks apply to AImedium

Before deploying an AI system that screens job applicants, a company wants to carry out a structured assessment of how the system could affect the people whose applications it processes, including potential harms and benefits to those individuals. Which ISO standard is purpose-built to guide that specific activity?

  • AISO/IEC 22989, which provides the foundational vocabulary and concepts used to describe artificial intelligence systems consistently across an organisation
  • BISO/IEC 42005, which provides guidance for conducting an artificial intelligence system impact assessment of effects on individuals and society Correct
  • CISO/IEC 42001, which sets out the requirements for establishing and maintaining an artificial intelligence management system across the organisation
  • DISO/IEC 23053, which describes a framework for artificial intelligence systems that are built using machine learning models and pipelines
Identify ISO/IEC 42005 as the standard guiding AI system impact assessment, distinct from the vocabulary, management system and engineering standards. ISO/IEC 42005 provides guidance on performing AI system impact assessments, helping an organisation document and analyse how a system can affect individuals and society throughout its lifecycle. That focus on assessing effects on affected people separates it from ISO 22989 (vocabulary), ISO 42001 (the management system) and ISO 23053 (machine learning system framework).

Why A is wrong: ISO/IEC 22989 is tempting because clear terminology underpins any assessment, but it defines AI concepts and vocabulary rather than telling an organisation how to assess a system's impact on affected people.

Why B is correct: ISO/IEC 42005 is the standard dedicated to AI system impact assessment, giving organisations a structured way to assess effects on individuals and society, which is exactly the task described.

Why C is wrong: ISO/IEC 42001 is plausible because an impact assessment is one activity within a management system, but the standard governs the overall system rather than providing the detailed method for a single impact assessment.

Why D is wrong: ISO/IEC 23053 is a believable choice because it is a recognised AI standard, but it frames how machine learning systems are engineered rather than how their impact on individuals is assessed.

Free sampleUnderstanding how laws, standards and frameworks apply to AIhard

A hospital signs up to a cloud vendor's AI service that transcribes and summarises clinical consultations. The hospital decides which consultations are recorded, why the summaries are produced and how long they are kept, while the vendor processes the audio strictly on the hospital's documented instructions. Under the GDPR, how are the two parties best characterised?

  • AThe hospital and the vendor are joint controllers, because the vendor's AI materially shapes the content of the clinical summaries it generates.
  • BThe vendor is the controller and the hospital is a processor, because the personal data is processed on the vendor's own infrastructure and platform.
  • CThe hospital is the controller and the vendor is a processor, because the hospital determines the purposes and means while the vendor acts on its instructions. Correct
  • DNeither party is a controller for the recordings, because the patients themselves decide whether their consultations are captured and summarised.
Distinguish controller from processor by who determines the purposes and means, not by who owns the infrastructure or builds the AI tool. The GDPR defines the controller as the party that determines the purposes and means of processing and the processor as the party that processes personal data on the controller's behalf and on its instructions. Because the hospital sets why, how and how long the data is processed and the vendor merely executes those instructions on its platform, the hospital is the controller and the AI vendor is its processor.

Why A is wrong: Joint control is tempting because the model clearly influences the output, but joint controllership turns on jointly determining purposes and means, not on the technical sophistication of a tool used under the controller's instructions.

Why B is wrong: Hosting the data on the vendor's platform can make it look like the vendor is in charge, yet controllership follows decision-making authority over purposes and means rather than ownership of the servers on which processing happens.

Why C is correct: The party that determines the purposes and means of processing is the controller, and a party that processes personal data only on the controller's documented instructions is a processor, which is exactly the split described here.

Why D is wrong: Obtaining patient consent is a lawful basis question and so seems relevant, but a data subject granting permission does not become the controller, and the hospital plainly still determines why and how the data is processed.

Other domains in this exam

See also the AIGP cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by IAPP. Original, blueprint-aligned practice material only.