Overview
Understanding how laws, standards and frameworks apply to AI has four competencies and 17 performance indicators, the most of any AIGP domain. II.A applies existing data privacy law: transparency, choice, lawful basis, purpose limitation, data minimization, privacy by design, controller obligations and special category data such as biometrics. II.B applies other existing laws: intellectual property, nondiscrimination in employment, credit, lending, housing and insurance, consumer protection against unfair and deceptive practices, and product liability. II.C covers the main elements of AI-specific law, with the EU AI Act as the leading example: risk categories, high-risk requirements, general-purpose AI models, enforcement and how duties differ for providers, deployers, importers and distributors. II.D covers the OECD AI Principles, the NIST AI Risk Management Framework and Playbook, and ISO standards 22989, 42001 and 42005.
IAPP allocates 19 to 23 scored items here, about 25% of the exam.
How it's tested
Domain II questions test applied legal reasoning rather than citation recall. A scenario describes an AI use, the parties and the data involved, and asks which law, risk tier, obligation or framework applies. Read for jurisdiction, role and type of data, because those three facts usually decide the answer.
-
Classify under the AI Act
A use case such as CV screening, emotion recognition or a chatbot is described, and you place it in the prohibited, high, limited or minimal risk tier.
-
Apply existing law
An AI system raises a privacy, copyright, discrimination or deceptive practice issue, and you identify the law and the obligation it triggers.
-
Match the framework
An organization wants a certifiable management system, a voluntary risk process or international policy principles, and you choose ISO/IEC 42001, the NIST AI RMF or the OECD Principles.
A European bank buys a vendor model to score consumer creditworthiness. Some candidates call the bank the provider because it uses the model on its customers. Under the AI Act the vendor that places the system on the market is the provider, while the bank is a deployer of an Annex III high-risk system. The bank therefore owes deployer duties such as human oversight, monitoring and a fundamental rights impact assessment.
Performance indicators 17
Official performance indicators from the IAPP outline, in outline order.
II.A.1 Understand how transparency, choice, lawful basis and purpose limitation requirements apply to AINotice, consent, lawful basis and purpose limits for AI.
- What it means
- This indicator applies core privacy principles to AI. Transparency requires telling people their data feeds an AI system and how. Choice covers consent and opt-outs. Lawful basis under the GDPR means each processing purpose, including training, needs a basis such as consent or legitimate interests. Purpose limitation restricts reusing data collected for one purpose to train models for another unless compatible.
- What to understand
- Training and inference are separate processing purposes and may need separate analysis. Scraped public data is still personal data under the GDPR. Legitimate interests requires a balancing test that considers people's reasonable expectations. Privacy notices must be updated before data is reused for AI. US state privacy laws add notice and opt-out rights for profiling in some states.
- How it shows up
- Customer data collected for service delivery is reused to train a model, and you identify the privacy issue and fix.
- Common confusion
- Assuming publicly available personal data needs no lawful basis for AI training.
II.A.2 Understand how data minimization and privacy by design requirements apply to AICollect only what is needed; build privacy in.
- What it means
- This indicator applies data minimization, which limits personal data to what is adequate, relevant and necessary, and privacy by design and by default, which requires building privacy protections into systems from the start (GDPR Article 25). AI's appetite for large data sets creates tension with minimization.
- What to understand
- Minimization is applied by justifying each feature, removing direct identifiers, using pseudonymization, aggregation or synthetic data, and limiting retention of prompts and outputs. Privacy by design adds techniques such as differential privacy, federated learning and access controls early, rather than retrofitting. Necessity must be argued for the purpose, not for model accuracy in general.
- How it shows up
- A team wants to collect every available attribute to improve accuracy, and you choose the minimization-compliant approach.
- Common confusion
- Thinking pseudonymized data is no longer personal data; it remains personal data under the GDPR.
II.A.3 Understand how obligations on data controllers apply to AIDPIAs, processors, transfers, data subject rights, automated decisions.
- What it means
- This indicator covers controller duties when AI processes personal data. IAPP's examples include privacy impact assessments, use of third-party processors, cross-border data transfers, data subject rights and automated decision-making. Under GDPR Article 35, likely high-risk processing such as systematic profiling needs a data protection impact assessment, and Article 22 restricts solely automated decisions with legal or similarly significant effects.
- What to understand
- Controllers decide purposes and means, so they remain accountable even when a vendor hosts the model. They need processor agreements, transfer mechanisms such as standard contractual clauses, and processes to honor access, correction, deletion and objection. Where Article 22 applies, people are entitled to human intervention, to express their view and to contest the decision, and must receive meaningful information about the logic involved.
- How it shows up
- An automated loan decision or vendor-hosted model is described and you identify the controller's obligation.
- Common confusion
- Assuming Article 22 bans all automated decisions; exceptions exist for contract, law and explicit consent with safeguards.
II.A.4 Understand the requirements that apply to sensitive or special categories of data (e.g., biometrics)Biometrics, health and other special category data.
- What it means
- This indicator covers stricter rules for sensitive data. GDPR Article 9 lists special categories such as health, biometric data for identification, racial or ethnic origin, religion, sexual orientation and political opinions, which need an explicit exception to process. US laws such as Illinois BIPA require notice and written consent before collecting biometric identifiers.
- What to understand
- AI raises two sensitive data issues: using such data directly, and inferring it from proxies such as postal code or purchase history. Facial recognition and voice prints are biometric when used to identify people. Some biometric AI uses are prohibited or high risk under the EU AI Act. Debiasing may require processing sensitive data, which the AI Act allows for high-risk systems under strict safeguards.
- How it shows up
- A facial recognition or health-inference use is proposed and you identify the heightened requirement.
- Common confusion
- Forgetting that inferring a sensitive trait can itself be processing of special category data.
II.B.1 Understand how intellectual property laws apply to AICopyright, training data, outputs and trade secrets.
- What it means
- This indicator covers IP questions on both sides of the model. On the input side, training on copyrighted works raises infringement claims, text and data mining exceptions and opt-outs, and licensing. On the output side, questions include whether AI-generated content can be protected and whether outputs reproduce protected works. Trade secret law protects model weights and confidential prompts.
- What to understand
- In the United States the Copyright Office position is that works need human authorship, so purely AI-generated material is not registrable. The EU allows text and data mining subject to rights holders' machine-readable opt-outs, and the AI Act requires general-purpose model providers to have a copyright policy and publish a training content summary. Contracts and acceptable use rules should address both inputs and outputs.
- How it shows up
- A model is trained on scraped content or produces near-copies of protected work, and you identify the IP risk.
- Common confusion
- Assuming content is free to use for training because it is publicly accessible online.
II.B.2 Understand how nondiscrimination laws apply to AIEmployment, credit, lending, housing and insurance.
- What it means
- This indicator applies anti-discrimination law to AI decisions in employment, credit, lending, housing and insurance. In the United States, examples include Title VII, the Equal Credit Opportunity Act, the Fair Housing Act and the ADA; the EU and UK have equality laws with similar aims. Disparate impact claims can arise even without intent to discriminate.
- What to understand
- Using AI does not change who is liable: the employer or lender remains responsible for discriminatory outcomes from a vendor tool. ECOA adverse action notices require specific reasons even when a complex model decides. New York City Local Law 144 requires bias audits and notice for automated employment decision tools. Removing protected attributes does not prevent proxy discrimination.
- How it shows up
- A screening or pricing model produces skewed outcomes for a protected group, and you identify the legal exposure.
- Common confusion
- Believing that excluding race or gender from inputs makes a model legally safe.
II.B.3 Understand how consumer protection laws apply to AIUnfair and deceptive acts or practices.
- What it means
- This indicator covers laws against unfair and deceptive acts or practices, such as Section 5 of the FTC Act and state UDAP laws. Regulators have applied them to exaggerated AI marketing claims, undisclosed use of AI, unfair use of biased systems and products trained on improperly obtained data.
- What to understand
- Deception means a misleading claim or omission that affects consumer decisions, such as advertising an AI capability that does not exist. Unfairness means substantial injury consumers cannot reasonably avoid that is not outweighed by benefits. The FTC has ordered companies to delete models trained on unlawfully collected data, a remedy often called algorithmic disgorgement. Claims about AI must be substantiated before they are made.
- How it shows up
- A company markets an AI feature or uses one on customers, and you identify the consumer protection risk.
- Common confusion
- Assuming consumer protection law does not apply until an AI-specific law is passed.
II.B.4 Understand how product liability laws apply to AIDesign and manufacturing defects, failure to warn.
- What it means
- This indicator applies product liability, which holds makers and sellers responsible for defective products, to AI. IAPP's examples include design and manufacturing defects. Failure to warn and negligence theories also arise. In the EU, the revised Product Liability Directive explicitly treats software, including AI, as a product.
- What to understand
- A design defect claim argues the AI was unreasonably dangerous as designed; a manufacturing defect claim argues a particular unit deviated from the design. Opacity makes proving defect and causation harder, which is why EU reforms ease the burden of proof in some cases. Documentation, testing records and warnings are the provider's best defense.
- How it shows up
- AI embedded in a product causes physical or financial harm, and you identify the liability theory.
- Common confusion
- Thinking software is never a product for liability purposes; the revised EU directive says it is.
II.C.1 Understand the risk classification framework for AI and what systems/uses fall into each categoryProhibited, high, limited and minimal risk tiers.
- What it means
- This indicator covers the tiered approach of AI-specific laws, led by the EU AI Act (Regulation (EU) 2024/1689). Prohibited practices include social scoring, manipulative techniques causing significant harm, untargeted scraping of facial images, emotion recognition in workplaces and schools, and most real-time remote biometric identification in public by police. High-risk systems are product safety components (Annex I) and listed uses (Annex III). Limited risk carries transparency duties; minimal risk carries none.
- What to understand
- Annex III areas include biometrics, critical infrastructure, education, employment, access to essential services such as credit and insurance, law enforcement, migration, and justice and democratic processes. An Annex III system can escape high-risk status if it performs only a narrow procedural or preparatory task and does not profile people. Prohibitions applied from 2 February 2025. The 2026 Digital Omnibus on AI added two prohibitions on generating intimate imagery or child abuse material, applying from 2 December 2026.
- How it shows up
- A use case is described and you place it in the correct risk tier.
- Common confusion
- Classifying all chatbots as high risk; most carry only transparency duties.
II.C.2 Understand the key requirements around risk management, data governance, technical documentation, conformity/impact assessments and record keepingHigh-risk system obligations, Articles 9 to 12 and 43.
- What it means
- This indicator covers what high-risk system providers must do under the EU AI Act. They need a risk management system across the life cycle (Article 9), data governance for training, validation and test sets including bias examination (Article 10), technical documentation (Article 11, Annex IV), automatic logging (Article 12) and a conformity assessment before market placement (Article 43), followed by CE marking and EU database registration.
- What to understand
- Most Annex III systems use internal control conformity assessment; some biometric systems and Annex I products involve a notified body. Certain deployers, including public bodies and those assessing credit or insurance, must carry out a fundamental rights impact assessment (Article 27). Under the Digital Omnibus, Annex III obligations apply from 2 December 2027 and Annex I product obligations from 2 August 2028.
- How it shows up
- A provider asks what it must complete before launching a high-risk system, and you choose the correct obligation.
- Common confusion
- Confusing the provider's conformity assessment with the deployer's fundamental rights impact assessment.
II.C.3 Understand the key requirements around human oversight, transparency and notification, and quality managementArticles 13, 14, 17 and 50.
- What it means
- This indicator covers further high-risk duties and AI Act transparency. Providers must design systems so humans can oversee them effectively (Article 14), supply instructions for use (Article 13) and run a quality management system (Article 17). Article 50 requires telling people they are interacting with an AI system, marking synthetic content in machine-readable form, labeling deepfakes and informing people exposed to emotion recognition or biometric categorization.
- What to understand
- Human oversight means people can understand limits, avoid automation bias, interpret outputs and override or stop the system. Deployers must assign competent overseers and inform workers before using high-risk AI at work. Article 50 obligations apply from 2 August 2026, with a short grace period to 2 December 2026 for machine-readable marking on generative systems already on the market.
- How it shows up
- A system interacts with the public or generates media, and you identify the transparency or oversight duty.
- Common confusion
- Thinking a human in the loop satisfies oversight even if that person cannot understand or override outputs.
II.C.4 Understand the distinct requirements for general-purpose AI modelsGPAI obligations and systemic risk models.
- What it means
- This indicator covers obligations for providers of general-purpose AI models, which apply from 2 August 2025. All GPAI providers must keep technical documentation, give information to downstream providers, have a policy to comply with EU copyright law and publish a sufficiently detailed summary of training content. Models with systemic risk, presumed above 10^25 floating point operations of training compute or designated by the Commission, face extra duties.
- What to understand
- Systemic risk duties include model evaluation and adversarial testing, assessing and mitigating systemic risks, serious incident reporting and adequate cybersecurity. Open-source GPAI models get partial exemptions unless they pose systemic risk. The General-Purpose AI Code of Practice offers a voluntary way to show compliance. Distinguish the model from the system: a chatbot built on a GPAI model is an AI system with its own tier.
- How it shows up
- A foundation model provider asks which duties apply, and you separate baseline GPAI from systemic risk obligations.
- Common confusion
- Assuming open-source release exempts a model from all AI Act duties, including systemic risk ones.
II.C.5 Understand the enforcement framework and penalties for noncomplianceMarket surveillance, AI Office and fines.
- What it means
- This indicator covers who enforces the EU AI Act and how. Member state market surveillance authorities supervise most AI systems, while the European AI Office in the Commission supervises general-purpose AI models. The European AI Board coordinates. Fines reach up to €35 million or 7% of worldwide annual turnover for prohibited practices, €15 million or 3% for most other breaches, and €7.5 million or 1% for supplying incorrect information.
- What to understand
- For each tier the higher of the two amounts applies, except for SMEs and start-ups, where the lower applies. Penalties for GPAI providers are imposed by the Commission. The Digital Omnibus gave the AI Office wider supervisory powers over certain GPAI-based systems and very large online platforms. Other laws, such as US state AI laws, use attorney general enforcement instead.
- How it shows up
- A breach is described and you identify the penalty tier or the authority that enforces it.
- Common confusion
- Mixing up the 7% tier for prohibited practices with the 3% tier for high-risk failures.
II.C.6 Understand the differences in requirements based on organizational contextProviders, deployers, importers and distributors.
- What it means
- This indicator covers how AI Act duties depend on role. Providers carry the heaviest load: conformity, documentation, quality management and post-market monitoring. Deployers use systems according to instructions, assign human oversight, monitor operation, keep logs and in some cases perform a fundamental rights impact assessment. Importers and distributors verify that systems carry required marking and documentation before passing them on.
- What to understand
- Roles can shift. A deployer, importer or distributor becomes a provider by putting its name or trademark on a high-risk system, making a substantial modification or changing its intended purpose so it becomes high risk. Authorized representatives act for non-EU providers. The Act also reaches providers outside the EU when outputs are used in the EU.
- How it shows up
- A company relabels or modifies a vendor system, and you determine whether its role and duties change.
- Common confusion
- Assuming roles are fixed by contract rather than by what the organization actually does with the system.
II.D.1 Understand the OECD principles, framework, policies and recommended practices for trustworthy AIOECD AI Principles and classification framework.
- What it means
- This indicator covers the OECD Recommendation on AI, adopted in 2019 and updated in 2024. It sets five values-based principles: inclusive growth, sustainable development and well-being; respect for the rule of law, human rights and democratic values including fairness and privacy; transparency and explainability; robustness, security and safety; and accountability. It adds five recommendations to governments, including investing in AI research and building human capacity.
- What to understand
- The OECD definition of an AI system, revised in 2023, shaped the EU AI Act's definition. The OECD Framework for the Classification of AI Systems groups systems by people and planet, economic context, data and input, AI model, and task and output. The principles are non-binding but widely adopted, and they underpin the G20 AI Principles.
- How it shows up
- A question asks which international principles or classification dimensions apply, and you pick the OECD element.
- Common confusion
- Treating the OECD Principles as binding law with penalties.
II.D.2 Understand the NIST AI Risk Management Framework and PlaybookGovern, Map, Measure, Manage and the Playbook.
- What it means
- This indicator covers NIST AI RMF 1.0 (January 2023), a voluntary US framework. Its core has four functions divided into categories and subcategories: Govern builds the culture, policies and accountability; Map establishes context and identifies risks; Measure analyzes and tracks risks with metrics; Manage prioritizes and responds to risks. The companion Playbook suggests actions for each subcategory.
- What to understand
- Govern is cross-cutting and informs the other three. NIST lists seven trustworthy characteristics: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair with harmful bias managed. Profiles adapt the framework to a context, such as the Generative AI Profile (NIST AI 600-1, 2024). It is not certifiable.
- How it shows up
- An activity such as inventorying context, running bias metrics or setting policy is described, and you identify the function.
- Common confusion
- Placing policy and accountability activities in Manage instead of Govern.
II.D.3 Understand the core ISO AI standards (i.e., 22989, 42001 and 42005)Terminology, AI management system and impact assessment.
- What it means
- This indicator names three ISO/IEC standards. ISO/IEC 22989:2022 sets AI concepts and terminology. ISO/IEC 42001:2023 specifies requirements for an AI management system (AIMS), following the harmonized management system structure with a Plan-Do-Check-Act cycle and Annex A controls; it is certifiable by accredited bodies. ISO/IEC 42005:2025 gives guidance on AI system impact assessment.
- What to understand
- Use 42001 when an organization wants an auditable, certifiable program that integrates with ISO 27001 or 27701. Use 42005 to structure assessments of impacts on individuals, groups and society. Use 22989 to align vocabulary. ISO/IEC 23894 on AI risk management is a related guidance standard, but the BoK names only these three as core. Certification to 42001 does not by itself show AI Act conformity.
- How it shows up
- An organization wants certification, common terms or an impact assessment method, and you pick the ISO standard.
- Common confusion
- Calling ISO/IEC 42005 certifiable; only 42001 specifies certifiable requirements.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
EU AI Act risk tiers
Prohibited, high, limited (transparency) and minimal risk.
- Remember
- Tier decides the obligations.
- Tested as
- Classification questions.
- Trap
- Treating limited risk as lightly regulated high risk.
-
Annex III high-risk areas
Biometrics, infrastructure, education, employment, essential services, law enforcement, migration, justice.
- Remember
- Now applies from 2 December 2027 after the Omnibus.
- Tested as
- Classification and timeline questions.
- Trap
- Using the original 2 August 2026 date without checking.
-
Provider vs. deployer
Places the system on the market versus uses it under its authority.
- Remember
- Rebranding or substantial modification makes a deployer a provider.
- Tested as
- Role questions.
- Trap
- Assigning by contract wording.
-
Conformity assessment
Provider's check that a high-risk system meets requirements before market placement.
- Remember
- Usually internal control for Annex III.
- Tested as
- Obligation questions.
- Trap
- Thinking every system needs a notified body.
-
Fundamental rights impact assessment
Deployer assessment under Article 27 for certain high-risk uses.
- Remember
- Public bodies, credit and insurance deployers.
- Tested as
- Deployer questions.
- Trap
- Assigning it to providers.
-
GPAI systemic risk
Models above 10^25 FLOPs or designated by the Commission.
- Remember
- Extra evaluation, incident and security duties.
- Tested as
- GPAI questions.
- Trap
- Thinking open source exempts systemic risk models.
-
AI Act penalties
€35M/7%, €15M/3%, €7.5M/1% tiers.
- Remember
- Higher amount applies; lower for SMEs.
- Tested as
- Enforcement questions.
- Trap
- Mixing the tiers.
-
GDPR Article 22
Limits solely automated decisions with legal or similar effects.
- Remember
- Right to human intervention and to contest.
- Tested as
- Privacy questions.
- Trap
- Reading it as a total ban.
-
Special category data
Health, biometrics for identification, race, religion and similar.
- Remember
- Inference can count as processing.
- Tested as
- Sensitive data questions.
- Trap
- Ignoring proxies.
-
Disparate impact
Neutral practice with disproportionate effect on a protected group.
- Remember
- Intent is not required.
- Tested as
- Nondiscrimination questions.
- Trap
- Assuming dropping protected fields is enough.
-
FTC Section 5
Prohibits unfair or deceptive acts or practices.
- Remember
- Applies to AI claims and data practices now.
- Tested as
- Consumer protection questions.
- Trap
- Waiting for an AI-specific law.
-
NIST AI RMF functions
Govern, Map, Measure, Manage.
- Remember
- Govern is cross-cutting.
- Tested as
- Framework questions.
- Trap
- Treating the RMF as certifiable.
-
ISO/IEC 42001
Certifiable AI management system standard.
- Remember
- Plan-Do-Check-Act with Annex A controls.
- Tested as
- Standards questions.
- Trap
- Equating certification with AI Act compliance.
-
OECD AI Principles
Five values-based principles plus five recommendations to governments.
- Remember
- Updated 2024; non-binding.
- Tested as
- Framework questions.
- Trap
- Treating them as enforceable law.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
NIST AI RMF vs. ISO/IEC 42001 vs. OECD AI Principles
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| NIST AI RMF | Voluntary US risk framework with Playbook | Process guidance, not certifiable | Selecting it when certification is required |
| ISO/IEC 42001 | International AI management system standard | Certifiable by third-party audit | Assuming it equals AI Act conformity |
| OECD AI Principles | Intergovernmental values-based principles | Policy level, non-binding | Using it as an operational control set |
EU AI Act risk tiers
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Prohibited | Unacceptable practices such as social scoring | Banned outright | Thinking a DPIA can make them lawful |
| High risk | Annex I products and Annex III uses | Full provider and deployer obligations | Missing the narrow-task exception |
| Limited risk | Chatbots, deepfakes, synthetic content | Transparency duties under Article 50 | Calling every chatbot high risk |
| Minimal risk | Spam filters, game AI | No mandatory duties | Adding high-risk controls by default |
AI Act roles
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Provider | Develops or has developed and places on the market | Conformity, documentation, QMS | Missing role shift by rebranding |
| Deployer | Uses under its authority | Oversight, monitoring, logs, FRIA if applicable | Assuming no duties |
| Importer | Places a non-EU provider's system on the EU market | Verifies conformity and marking | Confusing with distributor |
| Distributor | Makes available further down the supply chain | Checks marking and documentation | Assuming it must redo conformity |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Chatbot equals high risk
- What candidates pick
- Classify a customer service chatbot as high risk.
- Why it looks right
- It interacts with the public using AI.
- Why it's wrong
- Unless used for an Annex III purpose, a chatbot carries Article 50 transparency duties, not high-risk obligations.
Clue to watch for The system only answers questions and does not make Annex III decisions.
-
Buyer is the provider
- What candidates pick
- Call the bank that uses a vendor credit model the provider.
- Why it looks right
- The bank applies the model to its customers.
- Why it's wrong
- The vendor places the system on the market; the bank is a deployer unless it rebrands or substantially modifies it.
Clue to watch for The organization bought the system rather than building it.
-
Certifiable NIST
- What candidates pick
- Recommend NIST AI RMF certification to prove compliance to customers.
- Why it looks right
- NIST is a respected standards body.
- Why it's wrong
- The AI RMF is voluntary guidance with no certification scheme; ISO/IEC 42001 is the certifiable option.
Clue to watch for The scenario asks for third-party certification.
-
Pseudonymized is anonymous
- What candidates pick
- Treat pseudonymized training data as outside the GDPR.
- Why it looks right
- Direct identifiers have been removed.
- Why it's wrong
- Pseudonymized data can be re-linked with additional information and remains personal data subject to the GDPR.
Clue to watch for A key or mapping table still exists.
-
Drop protected attributes
- What candidates pick
- Remove race and gender fields to make a lending model compliant.
- Why it looks right
- The model can no longer see those traits.
- Why it's wrong
- Proxies such as postal code still produce disparate impact, so outcomes must be tested for bias.
Clue to watch for Outcomes still differ across groups.
-
Open source exemption
- What candidates pick
- Exempt an open-weight model above the systemic risk threshold from AI Act duties.
- Why it looks right
- Open-source models get some exemptions.
- Why it's wrong
- The exemption does not cover GPAI models with systemic risk, which keep evaluation, incident and security duties.
Clue to watch for Training compute above 10^25 FLOPs or Commission designation.
Answer framework
Work through these steps to identify the law, obligation or framework a Domain II scenario is testing.
- Identify the jurisdiction and the data involved, noting whether personal or special category data is used and whether training and inference are separate purposes.
- Check existing laws first: privacy, intellectual property, nondiscrimination, consumer protection and product liability all apply to AI without an AI-specific statute.
- If the EU AI Act applies, classify the system as prohibited, high, limited or minimal risk, and check whether a GPAI model is involved.
- Determine each organization's role (provider, deployer, importer, distributor) and list the obligations and enforcement exposure for that role and tier, using current Omnibus dates.
- Select the voluntary framework that matches the goal: OECD for principles, NIST AI RMF for risk process, ISO/IEC 42001 for certification, 42005 for impact assessment.
Practice question
Want more like this? The AIGP practice test has questions for every domain.
Checklist
0 of 16
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Split this domain into existing law, AI-specific law and frameworks, and study them in that order. For II.A, revisit GDPR Articles 5, 6, 9, 22, 25 and 35 and write one sentence on how each applies to training data and to automated outputs. For II.B, build a table of the four other legal areas with one US and one EU example for each.
The EU AI Act deserves the most time. Create a one-page map: risk tiers with examples, the Annex III list, provider and deployer duties, GPAI rules, penalty tiers and the timeline. Add the 2026 Digital Omnibus changes as a separate column, because exam items written before it may use the original dates. Practice role-shifting scenarios until you can spot a deployer becoming a provider.
For II.D, read the NIST AI RMF core and skim the Playbook, list the OECD principles, and read the scope sections of ISO/IEC 22989, 42001 and 42005. Then compare them in a three-column table by purpose, binding force and certifiability. Finish with practice questions and case studies that mix legal and framework questions.
FAQ
How much of the AIGP exam covers Understanding how laws, standards and frameworks apply to AI?
IAPP allocates 19 to 23 of the 85 scored items to Domain II, about 25% of the exam.
Does the AIGP exam focus only on the EU AI Act?
No. The EU AI Act is the main example of AI-specific law, but the domain also covers privacy, IP, nondiscrimination, consumer protection and product liability law, plus OECD, NIST and ISO frameworks.
Which EU AI Act dates should I study?
Prohibitions applied from February 2025 and GPAI duties from August 2025. The 2026 Digital Omnibus deferred Annex III high-risk duties to 2 December 2027 and product-embedded high-risk duties to 2 August 2028. Learn the original dates too, since BoK v2.1 predates the change.
Is the NIST AI RMF mandatory?
No. It is a voluntary framework from the US National Institute of Standards and Technology, organized into Govern, Map, Measure and Manage, and it has no certification scheme.
Which ISO standards does the BoK name?
ISO/IEC 22989 for concepts and terminology, ISO/IEC 42001 for AI management systems, and ISO/IEC 42005 for AI system impact assessment.
Do I need to memorize article numbers?
Knowing key articles such as GDPR Article 22 or AI Act Article 50 helps, but questions mostly test which obligation applies to which role and risk tier.
Are US state AI laws tested?
The BoK refers to AI-specific laws generally. Know the main state examples, such as Colorado's AI Act and New York City's bias audit law for hiring tools, as illustrations of risk-based duties.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 2 knowledge
You have covered privacy and other laws applied to AI, the EU AI Act and the OECD, NIST and ISO frameworks. Check whether you can classify systems and roles correctly under exam conditions.