Exam
AI governance certification: what should the exam test?
An AI governance certification tells an employer that a person passed an exam on how AI systems are registered, tested before launch, overseen after launch and bought from vendors. The IAE program in this field uses the NIST AI Risk Management Framework of January 26, 2023 and ISO/IEC 42001 of December 18, 2023 as models, and GDPR, CCPA/CPRA and the EU AI Act as reference frameworks. Each candidate gets their own figures and has to reach a decision.

What is an AI governance certification?
A credential issued to a person after an exam on managing the risks of AI systems. Certification under ISO/IEC 42001 is a different thing: it concerns an organization's AI management system.
According to the IEC webstore abstract, ISO/IEC 42001:2023 "specifies the requirements and provides guidance for establishing, implementing, maintaining and continually improving an AI (artificial intelligence) management system within the context of an organization." It applies to any organization that provides or uses products or services that utilize AI systems, regardless of size. The first edition was published on December 18, 2023 and runs to 51 pages; the committee is ISO/IEC JTC 1/SC 42, Artificial Intelligence.
A personal certificate says what one person knows and can do. It does not say that the person's employer complies with any law or standard. The IAE program in AI governance states this in its result section: the certificate "is not a license, a legal qualification or confirmation that an organization complies with the law."
Which documents does an AI governance exam draw on?
In the IAE program, GDPR, CCPA/CPRA and the EU AI Act are reference frameworks, and the NIST AI RMF and ISO/IEC 42001 are models of an AI risk management system.
NIST released AI RMF 1.0 on January 26, 2023 as "a guidance document for voluntary use by organizations designing, developing, deploying or using AI systems" (NIST news release). NIST prepared it on a direction from Congress in Public Law 116-283. Its core has four functions, govern, map, measure and manage, which NIST says "can be applied in context-specific use cases and at any stages of the AI life cycle." NIST also released a companion AI RMF Playbook with suggested ways to use the framework.
The EU AI Act is Regulation (EU) 2024/1689 of June 13, 2024. EUR-Lex already carries a version of it dated July 27, 2026. An exam that asks for article numbers and effective dates goes out of date with each amendment, and the IAE program excludes them: it tests how the principles of a framework apply to a situation.
ISO/IEC 42001 is written for organizations, and the IAE program lists certification audit procedures under management system standards among the topics it does not test. In the exam the standard is a model of an organization's AI management system.
What does the NIST AI RMF expect a professional to weigh?
Seven characteristics of trustworthy AI: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. NIST adds that trade-offs between them are usual.
Section 3 of the framework, published on the NIST AI Resource Center, lists the characteristics and draws accountable and transparent as a vertical box "because it relates to all other characteristics."
NIST does not expect all seven to apply in every system. The text says that "tradeoffs are usually involved, rarely do all characteristics apply in every setting," and gives an example: "organizations might face a tradeoff between predictive accuracy and interpretability."
The IAE program tests it with a task that requires a choice. Section S2 covers the choice of a decision threshold and "the trade-off between accuracy and the gap between groups." The candidate states which threshold to use and what the company gives up with it.
Should an AI governance exam include calculations?
Yes, when the decision depends on figures. Selection rates by group, error rates, the size of a sample check and notification deadlines decide whether a system launches, and IAE tasks give those figures.
For comparing selection rates between groups, U.S. employment regulations set a reference threshold. Under 29 CFR 1607.4(D), a selection rate for any race, sex or ethnic group "which is less than four-fifths (4/5) (or eighty percent)" of the rate for the group with the highest rate will generally be regarded by the federal enforcement agencies as evidence of adverse impact. Smaller differences may still count "where they are significant in both statistical and practical terms."
When a model screens CVs, the review before launch starts with these rates. The IAE case set includes a pre-launch review of an AI CV screener with given error rates by group, and section S2 lists selection rates and error rates by group, the ratio of rates and the four-fifths rule as a reference point.
Small groups make the rates uncertain. The program asks for the 95 % confidence interval of a rate and the sample size needed for a given precision, and it gives the formula in the preparation section. A calculator is enough for every calculation.
What does a good exam task look like?
A situation with the candidate's own figures and questions that end in a decision. The IAE sample asks how many false refusals face recognition at office turnstiles would produce in each group and whether to launch it.
In the published sample, an office has 1,200 employees, each passes a turnstile 4 times a day, and the vendor reports false refusals of 0.5 % for 900 employees with lighter skin tones and 2 % for 300 employees with darker skin tones. HR proposes to obtain consent in an addendum to the employment contract.
The rubric expects 18 false refusals a day in the first group and 24 in the second. Over 20 working days that is 0.4 refusals per employee in the first group and 1.6 in the second, four times as often. The second group is 25 % of the staff and receives 57 % of the refusals.
The same answer has to treat biometric data used for identification as a special category, explain why employee consent cannot be relied on without an equivalent alternative such as the badge, and name an impact assessment before launch, the storage and deletion of templates and a ban on the vendor training on the data. Both a refusal to launch and a voluntary mode with the badge kept are accepted if the conditions are stated.
How should the answers be scored?
Against a rubric published before the exam. IAE scores each task on correctness 40 %, reasoning 25 %, risks and constraints 20 % and clarity 15 %. A reference to a rule without applying it adds no points.
Correctness covers the calculation and the conclusion on the figures of the candidate's own task. Reasoning covers why one option was chosen over another. Risks and constraints cover consequences for people, data and vendors and what can go wrong after launch. Grammar and style do not lower the score if the meaning is clear.
A generic answer has a ceiling. In the question rubric, an answer whose calculation or conclusion does not rest on the candidate's own figures gets no more than 19 of 40 points for correctness. In the case rubric, a generic answer that does not choose between options also gets no more than 10 points for reasoning.
A reference to another framework or national law counts the same as a reference to GDPR, CCPA/CPRA or the EU AI Act if the requirement is stated correctly. The result is converted to 100 points using the section weights, and the pass mark is 70.
Should the exam test the law of one country?
The IAE exam does not. It uses GDPR, CCPA/CPRA and the EU AI Act as reference frameworks and accepts another framework or the candidate's own national law if the requirement is named and the decision justified.
The program says it in its first section: the exam "is not a legal qualification, does not give the right to practice law and does not test the law of any single country." Every candidate in the field is scored on the same rubric, wherever the candidate works.
The program excludes article numbers, exact wording and effective dates of rules, the law of a particular country and case law, legal drafting, writing code and training models, configuration of security tools, certification audit procedures, and detailed sector rules for medical devices, banking supervision of models and insurance.
Deadlines and thresholds that a calculation depends on, such as a breach notification deadline or a monitoring threshold, are given in the task, usually as company policy.
How does an online exam keep the answers the candidate's own?
IAE disables pasting, logs typing and tab switches, gives each candidate different figures, and adds a 5-minute written question when an answer is doubtful. Reference materials are allowed, AI-generated text is not.
Each task has its own timer: 10 minutes for a question and 25 minutes for a case. The exam runs in one sitting of up to 2 hours and cannot be paused once started. It is taken in English or Russian.
Under the exam terms a candidate may not hand tasks to third parties, use help from other people or submit text from AI services. Because every candidate gets different figures and circumstances in the cases, an answer copied from another candidate does not fit the task, and the rubric caps answers that do not work through the candidate's own numbers.
One question concerns a project from the candidate's questionnaire: an AI system or a data protection decision the candidate worked on personally, with the role, the decisions, the figures and the outcome.
What are the sections of the IAE exam in AI governance?
Five: AI inventory and risk classification 15 %, pre-launch assessment and testing 20 %, data protection 25 %, human oversight, monitoring and incidents 20 %, and vendors, contracts and governance practice 20 %.
The program IAE-EXAM-AI, version 1.0, took effect on November 2, 2026. The exam has 5 questions of 10 minutes, one for each section, 2 cases of 25 minutes and the project question of 10 minutes.
Section S1 covers the AI inventory and how to find unregistered services, risk classes under a reference framework, the roles of provider and deployer, controller and processor, and substantial changes to a model. S2 covers the impact assessment before launch, rates by group, uncertainty in small groups, the decision threshold and the launch decision.
S3 covers the lawful basis for processing, minimisation, retention periods, data subject requests including deletion of trained models, and cross-border transfer. S4 asks what real human oversight looks like, for example time per decision and signs that reviewers agree with the model without checking, and covers drift monitoring, the decision to stop a model and breach timelines. S5 covers vendor checks and contract terms: training on customer data, model version changes, sub-processors, incident notification, the right to audit and exit.
Who can take the IAE exam, and what does the certificate state?
A candidate with a bachelor's degree or higher and 3 years of experience in the field, or 5 years without a degree. A score of 70 or more gives an IAE Certified Professional certificate for 3 years.
The candidate fills in a 15-minute questionnaire on the exam page and pays by card: $300, or $200 for IAE members. The Secretariat decides on admission within 1 business day and refunds the fee in full if the questionnaire does not meet the requirement. The exam is taken on any of the 14 days after admission. IAE membership is not required.
Answers are scored against the field rubric, an examiner approves the result, and the result and a score report by section arrive by email within 5 business days. The certificate and the public IAE registry show the field AI Governance and Data Protection, the program IAE-EXAM-AI and its version, the certificate number and the validity date.
A retake is possible 30 days after the first failed attempt and 60 days after the second, with no more than 3 attempts in 12 months. An appeal is filed within 30 days of the result and is reviewed by a Council member who did not score the work, with a decision within 10 business days. The Credentialing Council reviews the program at least once a year. IAE does not promise that the certificate will affect a USCIS decision or an employer's hiring decision.
Sources
- NIST, NIST Risk Management Framework Aims to Improve Trustworthiness of Artificial Intelligence (January 26, 2023)
- NIST AI Resource Center, AI RMF section 3, AI Risks and Trustworthiness
- IEC Webstore, ISO/IEC 42001:2023, Artificial intelligence management system
- EUR-Lex, Regulation (EU) 2024/1689 (Artificial Intelligence Act)
- 29 CFR 1607.4, Information on impact (LII)
Read also
All articles
How many times can you retake a certification exam, and how long do you wait between attempts?
Every certification body writes its own retake rule. AWS and The Open Group set no cap on attempts, Microsoft permits five in 12 months, PMI three
Do professional certifications expire, and how does recertification work?
Most professional certifications are issued for a fixed period, often three years, and renewed through continuing education or a new exam. ASQ, NICET
Can a professional certification count as degree equivalency for an H-1B?
Yes, in one narrow way. Under 8 CFR 214.2(h)(4)(iii)(D)(4), certification or registration from a nationally-recognized professional association
Book the exam
The questionnaire takes 15 minutes. Admission decision in 1 business day, exam on any of 14 days.
