AI governance training for risk, legal and audit teams: a curriculum
A practitioner curriculum for risk, legal, compliance and audit teams: six modules, case exercises set in KSA, UAE and Jordan, and an honest way to assess.
Your risk, legal, compliance and internal audit teams will be asked to approve AI assistants and review AI vendors. They will also rate AI risks and investigate when an AI tool exposes customer data. A general AI course teaches them to write prompts; it does not teach them to do those jobs. AI governance training for these teams needs its own curriculum, with each topic practised on a case. It covers AI risk in their vocabulary, data protection in Saudi Arabia, the UAE and Jordan, ISO/IEC 42001 basics, impact assessment, agent and vendor reviews, and incidents.
Who needs AI governance training, and why a general AI course is not enough
Staff-wide training, such as our role-based AI training for employees, helps people use approved tools well. Control functions have a different job. They do not need to be the best users of Claude or Microsoft Copilot. They need to judge other people's use of them and write findings that hold up.
A compliance officer asked to clear a call-centre chatbot needs to know which data it sees, where it is processed, who answers for its mistakes and what evidence to keep. A general course covers none of that.
| Role | What they should be able to do after the training |
|---|---|
| Risk management | Rate an AI use case on the organisation's scale and record it in the AI risk register |
| Legal and data protection (including the data protection officer, where one is appointed) | Identify which data protection law applies, spot lawful-basis and transfer questions, and frame precise questions for counsel |
| Compliance | Check a proposed use against the AI policy and sector rules, and decide what must be escalated |
| Internal audit | Plan and test AI controls, and write findings backed by evidence |
| IT security and procurement | Review an AI vendor's or agent's permissions, data handling and contract terms |
We recommend a cohort of 10 to 15 people that mixes these roles and includes one or two product or workflow owners to keep the controls workable.
The curriculum: six modules and a capstone
Each module runs for about three hours: a short input, a template to fill in and a case exercise on a fictional organisation. Spread over three to six weeks, the gaps let people try each template on their own work.
| Module | What participants learn and practise | Case exercise |
|---|---|---|
| 1. AI risk foundations | How AI risks differ from IT risks: confident wrong answers, data leaking through prompts, bias, over-reliance; rating likelihood and impact | Rate five AI use cases for a fictional insurer and enter them in a risk register |
| 2. Data protection in KSA, UAE and Jordan | Which law applies, personal and sensitive data, lawful bases, transfers outside the country, breach duties | Map the data flows of a customer chatbot that serves two countries |
| 3. ISO/IEC 42001 basics | What an AI management system is and how its parts connect: policy, roles, risk and impact assessment, monitoring, improvement | Trace one AI system through the management system and list the missing evidence |
| 4. AI impact assessment | Who a system can affect, how to describe harms, and what should change a go-ahead decision | Draft an impact assessment for a CV screening tool |
| 5. Agent and vendor reviews | An agent's tools, data access and human confirmation points; vendor questions on data use, retention, hosting and sub-processors | Review a vendor's answers and an agent's permissions; accept, accept with conditions or reject |
| 6. AI incidents | What counts as an AI incident, who decides, what to log, when data breach rules apply | A timed tabletop on a chatbot that showed one customer's data to another |
| Capstone | A full review of one fictional AI system, using every template above | Written submission, scored on the rubric below |
Module 2: data protection in general terms
The module teaches the shape of each law, not its detail. Saudi Arabia's Personal Data Protection Law was issued by Royal Decree No. (M/19) dated 9/2/1443H and amended by Royal Decree No. (M/148) dated 5/9/1444H, as cited in the SAMA Rulebook. In the UAE, Federal Decree-Law No. 45 of 2021 on personal data protection came into force on 2 January 2022, according to the UAE government portal. It does not cover every case: free zones with their own law, such as the DIFC, and sectors with their own legislation, such as health, follow separate rules. Ask counsel whether its executive regulations have been issued and what they require.
Jordan's Personal Data Protection Law No. (24) of 2023 requires a data protection officer in listed cases, including processing sensitive data and transferring databases outside the Kingdom (Article 11 of the official English translation).
This module is general information, not legal advice: obligations depend on the organisation's sector, data and countries of operation, and sector regulators can add rules. Ask your own counsel to review the module before it runs.
Modules 3 and 4: the two ISO standards
ISO/IEC 42001:2023, published in December 2023, specifies requirements for establishing, implementing, maintaining and continually improving an AI management system. ISO describes it as applying to organisations of any size that develop, provide or use AI, including public sector agencies. ISO/IEC 42005:2025, published in May 2025, gives guidance on AI system impact assessments: how a system and its foreseeable uses may affect individuals, groups or society across its lifecycle (ISO). Both modules teach the standards by structure and purpose, not their text. On certification itself, see our article on ISO 42001 certification readiness.
Module 6: incidents against a real clock
Incident exercises need a clock, and Jordan's law supplies one in Article 20. Where a breach of data security is likely to cause severe harm to data subjects, the controller notifies them within 24 hours of discovering it. It also reports to the regulatory unit at the Ministry of Digital Economy and Entrepreneurship within 72 hours. The team decides, inside those windows, whether an AI output that exposed customer data meets that test, and who signs the report.
Two case exercises set in the region (illustrative)
Both cases below are fictional and illustrative.
A Saudi bank's internal audit team plans its first AI review
A mid-sized bank in Riyadh runs a call-centre AI assistant and a vendor tool that summarises credit files. A SAMA circular dated 19/5/1443H (23 December 2021), published in its Rulebook, asks supervised institutions to assess their gaps against the Personal Data Protection Law and the data governance policies of SDAIA and NDMO. It also asks them to present a plan to close those gaps to the board for approval. Auditors write a one-page audit plan for both tools, linked to that gap plan, with the evidence they would request and the first two controls to test. The facilitator's answer key rewards testable controls, not statements such as "AI is used responsibly".
A Jordanian insurer's legal team reviews a claims chatbot vendor
A family-owned insurer in Amman wants a claims chatbot hosted outside Jordan. Article 15 of Jordan's law bars transfers to a recipient whose level of protection is lower than the law's, except in listed cases. The controller must verify that level before the transfer starts. Participants list what the vendor must answer in writing (training on customer data, retention, hosting, sub-processors, breach notice) and decide whether the deal can proceed with conditions. A strong answer separates what the law requires from the insurer's own risk appetite.
How to assess practitioners without pretending to certify
We do not certify anyone. ISO/IEC 42001 certification applies to an organisation's management system and is granted by independent certification bodies. What a training program can honestly give is a training record: attendance, the work submitted and the internal assessment scores.
We recommend scoring filled templates rather than a multiple-choice test. Someone who can name a law but cannot rate a use case has not learned the job. A four-point rubric is enough:
| Criterion | 1: not yet | 4: strong |
|---|---|---|
| Risk judgement | Ratings without reasons | Ratings with reasons, consistent with the register scale |
| Use of law and standards | General statements | Points to the right rule and flags what needs counsel |
| Evidence | Asks for "a policy" | Names the document, log or test that shows the control works |
| Decision | No clear outcome | Accept, accept with conditions or reject, with owner and date |
A fair completion rule: at least 80% attendance, an average of 3 or more, and a capstone scored 3 or more. Then check again after 60 days: new entries in the AI register, impact assessments started, vendor reviews written. That shows whether the training changed the work; attendance does not.
Train before or after you build the governance framework?
We recommend training alongside the first governance documents, not before them. Exercises built on the organisation's own AI policy, risk scale and register template give people something to apply. The trade-off is time: waiting for a finished framework can delay training by months, so start once a draft policy and a register template exist.
Our AI Governance & Compliance Foundation (GB4) covers policy, decision rights, AI and risk registers, PDPL alignment and ISO/IEC 42001 readiness. AI governance training then prepares the people who will run them. If a regulator question started the discussion, see our page on PDPL and regulators.
How to start
In our Academy Lab, this curriculum fits Role-Based AI Training (AH2): hands-on training on the team's real tasks, with a track for managers and workflow owners. For control teams, those tasks are reviews, assessments and findings; fictional cases stand in where real files are too sensitive. If you are not sure where your gaps are, the AI Exposure Scan is a free 45-minute conversation that ends in a written next step.
Frequently asked questions
Is AI governance training the same as ISO/IEC 42001 certification?
No. ISO/IEC 42001 certification applies to an organisation's AI management system and is granted by independent certification bodies after an audit. Training builds the skills of the people who run governance: rating risks, assessing impact, reviewing vendors and handling incidents. The outline we recommend ends in a training record of attendance and internal assessment scores; it does not award a certificate or a professional credential.
Who should attend AI governance training?
The people who approve, review and audit AI use: risk managers, legal and data protection staff, compliance officers, internal auditors, and IT security and procurement staff who review vendors. Adding one or two product or workflow owners keeps the discussion practical. We recommend a mixed cohort of 10 to 15 people, so each case is argued from several sides, as in an approval meeting.
How long does an AI governance training program take?
The outline we recommend has six modules of about three hours each plus a capstone review, spread over three to six weeks. The gaps between sessions matter: participants try each template on their own work before the next module. A follow-up check after 60 days looks at whether the templates are in use, such as new AI register entries and impact assessments started.
Do risk and audit staff need technical AI knowledge for this training?
They need enough to ask the right questions, not to build models. The first module explains how AI systems fail in ways ordinary software does not, such as confident wrong answers or data leaving through prompts. Later modules focus on judgement: which law applies, what evidence proves a control works, and when an agent's actions need human confirmation before they take effect.
Get new insights by email
One email when we publish something useful. No spam; leave any time.
Services in this article
Related lab: Academy Lab ↗
