Get consulting from Sterling
Expert and experienced consultants, 8,100+ client engagements in 60+ countries, from gap assessment to a system ready for its certification audit.
Talk to a Sterling consultantPractitioner training · AI governance
Online training course to run AI risk management as a working programme, across every AI system you build or use.
This practitioner training course shows how an organisation runs AI risk management as an ongoing programme: AI risk in enterprise risk management, risk appetite and the three lines, the ISO/IEC 23894 process, the NIST AI RMF and its Generative AI Profile, the EU AI Act risk management system, deployer duties and fundamental rights impact assessment, suppliers, testing, monitoring, incidents and board reporting. A consumer bank's credit model and customer assistant run through the course.
Secure payment by Stripe · for yourself or for your team
Add another course and save 15%, up to 25%
What you get
About 3½ hours, with captions. Watch at your own pace on any device.
Short questions check your understanding before you move on.
Score at least 60% to receive the Certificate of Competence.
Industry experts, domain specialist trainers and globally renowned auditors with 30+ years in ISO and GRC.

Verifiable certificate with a QR code anyone can scan, plus a digital badge to add to your LinkedIn profile.
See how verification works →Save when you take more than one course
The discount is added automatically in your basket.
Standard and editionISO/IEC 23894:2023, NIST AI RMF 1.0 and the EU AI Act
Last updated11 November 2026
Preview this course
Watch a sample from AI Risk Management Training (ISO/IEC 23894, NIST AI RMF and EU AI Act) before you enrol.
What learners say
Every course is built by the experts behind 8,100+ client engagements in 60+ countries, so you learn what auditors and regulators actually expect. Here is what professionals said after finishing their course.
Informative
10/10 for Philippines Data Privacy Act and DPO Training
Best value
All-Access Pass
A pharma and life sciences professional from the USA started with one course, then upgraded to the All-Access Pass.
Every Sterling Academy course for 12 months.
US$499
See the passA selection of comments from learners who finished a course and answered our feedback request, shortened from their own words. Client names available on request.
Learning outcomes
Connect the subject to the responsibilities and decisions people encounter in their roles.
Part of: AI Governance and ISO 42001 Training
Inside the course
14 modules, 14 practice checkpoints, 4 practice exercises and a final exam. Video runtimes are approximate; allow additional time for the practice questions and the final exam. Modules marked optional can be skipped; the others are studied in any order.
Why AI risk management matters: three real AI cases, why AI risk differs from other risk, the four sources the course combines and key EU AI Act dates.
Opens the AI Risk Management training course with three cases as the regulators and tribunal described them: a credit card investigated by New York's Department of Financial Services, a pharmacy chain's facial recognition in a proposed FTC order, and an airline chatbot decision in British Columbia. Draws three lessons: keep the evidence, test and keep watching, and own what your AI does. Explains why AI needs its own approach, using the NIST AI Risk Management Framework, and the two definitions of risk from ISO 31000 and the EU AI Act. Introduces the four sources, the EU AI Act dates as amended in 2026, and KVB Consumer Bank in Frankfurt with its credit-decision model and customer assistant, then the three parts of the course.
What you will be able to do
Why it matters
Regulators and tribunals have shown that you own what your AI does, so evidence, testing and watching matter from day one.
Practice: Practice: Why AI risk management matters · 3 questions
Placing AI risk inside enterprise risk management: risk appetite and tolerances, risk owners, the Three Lines Model, the AI inventory and decision rights.
Begins with KVB's board risk committee asking for its AI risk appetite and finding nobody could answer. Covers ISO 31000 clause 5.2 and four of its principles, NIST's view that AI risk is integrated, and the difference between appetite, tolerance and risk criteria. Shows KVB's four appetite statements and the tolerances with early triggers that make them measurable, both built for the course. Explains risk owners under ISO 31000 and ISO/IEC 42001, NIST's GOVERN outcomes, and the IIA's Three Lines Model applied to KVB's credit model. Builds the AI inventory under GOVERN 1.6, sets decision rights for residual risk by level, and gives examples from an insurer in Paris and a hospital group in London.
What you will be able to do
Why it matters
An AI committee with its own scales hides AI risk from the board; one register lets it be weighed against credit or cyber risk.
Practice: Practice: AI risk in the enterprise · 3 questions
How ISO/IEC 23894 guidance on AI risk management works with ISO 31000: its structure, the principles it extends for AI, and the risk management framework.
Starts with KVB's risk team deciding whether to write a separate AI method or adapt its ISO 31000 framework. Explains what ISO/IEC 23894 is, its scope and its map: clause 4 on principles, clause 5 on the framework, clause 6 on the process and three informative annexes. Covers the warning about different meanings of risk and the five principles it extends for AI: inclusive, dynamic, information, human and cultural factors, and continual improvement. Works through clause 5, including the suggested commitment statement KVB now publishes, the external and internal context in Tables 2 and 3, and named roles under clause 5.4.3, with examples from Singapore and Sydney.
What you will be able to do
Why it matters
Two registers give two answers about the same system; extending the framework you have keeps one picture for the board.
Practice: Practice: The ISO/IEC 23894 approach · 3 questions
Setting the scope, context and AI risk criteria for one AI system using clause 6 of ISO/IEC 23894 and the objectives in its first annex.
Opens with KVB's first AI risk workshop, which produced two hundred sticky notes and no decisions. Explains clause 6.1 and risk processes at four levels, the wider stakeholder list in clause 6.3.3 and the context considerations it lists. Shows the scope statement for KVB's credit model under clauses 6.3.1 and 6.3.2, and the AI-related objectives in the standard's first annex, of which KVB chose six. Covers Table 4's four points for AI risk criteria, the uncertainty hidden in past lending decisions, and KVB's consequence scale described for the bank, individuals and society. Compares a robot cell in Tokyo and a benefit triage tool in Amsterdam.
What you will be able to do
Why it matters
If the consequence scale cannot see people, the register will not either; agreeing criteria first makes the risk list usable.
Practice: Practice: Context and risk criteria · 3 questions
Identifying AI risks with ISO/IEC 23894: assets, the ten areas of risk sources and the annex catalogue, events, controls that really work and consequences.
Starts with KVB's first list of credit model risks, which held only bias and hacking. Explains clause 6.4.1 and how KVB matched assessment methods to life cycle stages. Works through the five parts of clause 6.4.2: assets for the organisation, individuals and society; risk sources across ten areas, with the standard's sample catalogue; events found through incident reports, field trials and interviews; controls rated by how they operate in practice; and consequences, including who gains and who bears the harm. Shows the entries the structured search added to KVB's register, such as the hand-over to credit officers and lost scorecard know-how, with examples from London and Dublin.
What you will be able to do
Why it matters
A brainstorm finds only familiar risks; a structured search finds the ones nobody has met before.
Practice: Practice: Identifying AI risks · 3 questions
Analysing, evaluating and treating AI risks with ISO/IEC 23894: three impact analyses, likelihood, treatment options, verification and the risk record.
Opens with a scoring meeting at KVB that rated harm to applicants low because impact was only in euros. Explains clause 6.4.3's business, individual and societal impact analyses and its caution on likelihood. Analyses KVB's thin-file risk from each viewpoint, evaluates it against the criteria and walks through the treatment options in clause 6.5.2, the risk-benefit analysis and verification under clause 6.5.3, where the measured ratio rose from 0.81 to 0.90. Treats a second risk, officers signing declines without real review. Covers the risk record in clause 6.7 and its feedback loop, with examples from Hong Kong and Madrid, and introduces a practice exercise.
What you will be able to do
Why it matters
A treatment marked done on trust changes nothing; changing the register only when the evidence changes keeps it reliable.
Practice: Practice: Analyse, evaluate, treat · 3 questions
The NIST AI Risk Management Framework 1.0: the seven characteristics of trustworthy AI, the Govern, Map, Measure and Manage functions, and profiles.
Begins with a partner bank's questionnaire that exposed gaps in how KVB's practices fitted together. Explains NIST AI 100-1, its voluntary use and its two parts, then the seven characteristics of trustworthy AI and the trade-offs between them. Shows how the four functions work, with Govern running through the others and never used as a checklist, and walks through each function's categories with KVB's records for the credit model. Explains current and target profiles, gives a mapping to the ISO process built for the course, and shows a Toronto retailer and a Singapore logistics firm starting in different places. KVB's profile found two gaps: an appeal route and a plan to switch the model off.
What you will be able to do
Why it matters
Marking every subcategory done says nothing about real outcomes; testing each against evidence shows what is missing.
Practice: Practice: NIST AI RMF 1.0: four functions · 3 questions
Generative AI risk using NIST AI 600-1, the Generative AI Profile: its twelve risks, confabulation, prompt injection, red-teaming and a release gate.
Starts with KVB's customer assistant pilot inventing a fee, revealing its rules and being trusted by staff without checks. Explains NIST AI 600-1 and lists its twelve risks. Looks closely at confabulation, prompt injection and data poisoning, and human-AI configuration, with the transparency duty in Article 50(1) of the EU AI Act. Recalls the British Columbia chatbot decision and the controls KVB drew from it, then KVB's list of five risks that apply. Covers the four primary considerations, the four types of red-teaming and KVB's four-step release gate, with examples from a Madrid telecom operator and a Boston healthcare start-up.
What you will be able to do
Why it matters
A generative system changes with each model version and document update, so its testing must keep pace or customers find the errors.
Practice: Practice: Generative AI risk: NIST AI 600-1 · 3 questions
The EU AI Act for high-risk AI: provider and deployer roles, classification under Annex III, the Article 9 risk management system and the amended dates.
Opens with KVB's general counsel asking whether the credit model is high-risk and whether KVB is its provider or deployer. Explains both roles under Article 3, classification under Article 6(2) and Annex III, point 5(b), the Article 6(3) derogation and the profiling rule, and shows KVB's signed classification record. Lists the Section 2 requirements, then works through Article 9 paragraph by paragraph: the four-step process, the order for choosing measures, testing against thresholds fixed in advance, vulnerable groups and combining with existing risk processes. Gives the dates as amended by Regulation (EU) 2026/1744, with examples from Dublin and Amsterdam.
What you will be able to do
Why it matters
Article 9 asks for a process that runs for the whole life cycle, not a report that sits in a file.
Practice: Practice: EU AI Act: Article 9 · 3 questions
Deployer duties under Article 26 of the EU AI Act, the right to an explanation in Article 86, and the fundamental rights impact assessment in Article 27.
Starts with KVB's credit officers agreeing with the model in almost every case after reviews of about forty seconds. Explains the Article 26 duties: following instructions, human oversight by competent people with authority, input data, monitoring, logs and informing workers and affected people. Covers the right to an explanation under Article 86 and KVB's deployer record, and how KVB splits its provider and deployer roles. Walks through the fundamental rights impact assessment in Article 27, including the paragraphs replaced by Regulation (EU) 2026/1744, ISO/IEC 42005:2025, and KVB's main entries. Gives the dates and examples from Amsterdam and Paris, with a practice exercise.
What you will be able to do
Why it matters
Oversight is a skill with authority, not a signature; the deployer carries these duties even when it did not build the system.
Practice: Practice: Deployer duties and the FRIA · 3 questions
Third-party and supplier AI risk: the models, data and services you buy, Article 25 of the EU AI Act, due diligence, contracts and monitoring.
Opens with KVB's model supplier retiring a model version at two weeks' notice, after which wrong answers on fees appeared. Covers ISO/IEC 42001 controls A.10.2 and A.10.3, NIST's GOVERN 6, MAP 4 and MANAGE 3 outcomes, and the value chain risk in the Generative AI Profile. Explains Article 25 of the EU AI Act, including when a deployer becomes a provider and the replaced Article 25(4). Shows KVB's register of four AI-related suppliers, five due diligence questions given as Sterling's recommended practice and a rewritten contract schedule, then recalls the FTC allegations about vendors and sets out supplier monitoring, with examples from Hong Kong and Tokyo.
What you will be able to do
Why it matters
You can buy the model, but you cannot hand over the risk; a supplier's change becomes yours unless contracts and monitoring catch it.
Practice: Practice: Third-party and supplier AI risk · 3 questions
Testing AI systems before release, monitoring in use, data drift, serious AI incidents under the EU AI Act and drilling the plan to switch a model off.
Starts with KVB's fairness ratio for young applicants sliding below its review trigger while it was only checked each quarter. Explains why AI must be watched in use, then testing before release under ISO/IEC 42001 and NIST, and monitoring in use. Shows KVB's rewritten monitoring plan with fast and slow signals, and the FTC allegations about missing monitoring. Covers post-market monitoring in Article 72, the definitions of serious incident and AI incident, and reporting under Article 73. Sets out KVB's five-step incident process and its switch-off drill, which improved from two days to target, with examples from Mumbai and Sydney.
What you will be able to do
Why it matters
A model can pass every test at launch and still drift as the world changes; a plan that is only written is only a hope.
Practice: Practice: Testing, monitoring and AI incidents · 3 questions
Metrics and board reporting for AI risk: risk indicators against appetite, programme health metrics, five numbers, fixed status rules and decisions first.
Begins with KVB's thirty-page report that could not tell the board whether it was within appetite. Explains ISO 31000 clause 6.7, ISO/IEC 42001 clauses 9.1 and 9.3, NIST outcomes and the IIA's view. Distinguishes risk indicators from programme health metrics and shows KVB's front page of five numbers, its green, amber and red rules fixed in advance and the decisions the committee is asked to take. Covers residual risk acceptances with end dates, NIST's warning about metrics that mislead, reporting to four audiences and the role of internal audit, with questions the committee asks and examples from Paris and Madrid.
What you will be able to do
Why it matters
A data dump gets noted and forgotten; a decision paper lets a board act on AI risk in the time it has.
Practice: Practice: Metrics and board reporting · 3 questions
Bringing the AI risk programme together: the ISO/IEC 23894 life cycle map, the programme on one page, the audit view, a ninety-day road map and the exam.
Closes the training course with KVB's first internal audit of AI risk management, one year on. Walks through the standard's third annex, which maps the risk process from inception to retirement, and the credit model's life cycle with its evidence. Shows KVB's whole programme on one page and how it gives a base for ISO/IEC 42001 certification. Sets out an auditor's checklist, a ninety-day road map and yearly rhythm as Sterling's recommended practice, and the EU AI Act dates as amended. Compares treating AI risk as a project or as a programme, with examples from Toronto and Boston, then gives five takeaways and explains the final exam.
What you will be able to do
Why it matters
AI changes in use, so a one-off assessment soon describes a system that no longer exists; a running programme keeps the record true.
Practice: Practice: Putting it together · 3 questions
About this course
AI systems now decide who gets credit, answer customers in the company's name and run inside products bought from suppliers. The cases in this course show what happens when their risks are not managed as a programme. In March 2021, New York's financial regulator found no evidence of discrimination in a bank's credit card decisions, but found weaknesses in transparency and customer service; the bank was able to identify the factors behind each decision it was asked about. In December 2023, the US Federal Trade Commission alleged that a retailer used facial recognition without testing its accuracy, monitoring it after launch or overseeing its suppliers. In February 2024, a tribunal in British Columbia held an airline responsible for what its website chatbot told a customer. The law is catching up: under the EU AI Act, as amended by Regulation (EU) 2026/1744, the high-risk duties for systems such as credit scoring apply from 2 December 2027.
The course shows how an organisation runs AI risk management as an ongoing programme, whether or not it seeks ISO/IEC 42001 certification. It places AI risk inside enterprise risk management with ISO 31000:2018: a written appetite, measurable tolerances with early triggers, named risk owners, the three lines and an AI inventory. It applies ISO/IEC 23894:2023, the international guidance that extends ISO 31000 for AI, step by step to one AI system: scope and risk criteria, the standard's catalogues of AI objectives and risk sources, impact on the business, on individuals and on society, treatment that is verified, a risk record that traces each risk, and the standard's map of the process across the AI life cycle. It uses the NIST AI Risk Management Framework 1.0 and its Generative AI Profile, NIST AI 600-1, and teaches the EU AI Act risk duties from the official text: the Article 9 risk management system, the deployer duties of Article 26, the fundamental rights impact assessment of Article 27, post-market monitoring and serious incident reporting. One consumer bank in Frankfurt runs through every module with two AI systems, a credit-decision model and a generative AI customer assistant, and its records are built on screen. Four practice exercises with model answers let you build the same records for your own systems.
The course is written for the people who own, check and report AI risk: AI risk and governance leads, risk managers and chief risk officers who run the programme; model risk, validation and data science teams who test and fix the systems; compliance, privacy and legal staff who own the legal duties and the impact assessments; and internal auditors who give the board independent assurance. Product owners and business leaders who sponsor AI systems will find their part too. It suits organisations that build their own models, those that buy AI from suppliers, and those that do both. No prior knowledge of the NIST framework or the EU AI Act is needed. Those who want the ISO/IEC 42001 risk and impact assessments for certification, the NIST framework in full depth, or the board's view of AI governance will find separate Sterling Academy courses on each.
After the course, you can write an AI risk appetite with tolerances and triggers, assign risk owners and decision rights for residual risk, and keep an AI inventory. You can build a NIST profile for an AI system, find its gaps and turn them into a plan, and decide which of the twelve generative AI risks apply to a system and how to test each one. You can classify a system under the EU AI Act, decide whether your organisation is its provider, its deployer or both, run an Article 9 risk management system, and draft a fundamental rights impact assessment. You can set due diligence questions and contract terms for AI suppliers, a monitoring plan with fast and slow signals, an AI incident process with its reporting clocks, and a switch-off drill. And you can write a board report that leads with decisions. The scenario-based final exam leads to the Sterling Academy Certificate of Competence for those who score 60 percent or more, with one retake included. It is a training certificate, not a certification of your organisation.
Frequently asked questions
About AI Risk Management Training (ISO/IEC 23894, NIST AI RMF and EU AI Act), its certificate and enrolment.
Ask SterlingYes. Sterling Academy's AI Risk Management Training (ISO/IEC 23894, NIST AI RMF and EU AI Act) training is built by the Sterling consulting group, which has taken organisations to certification across 8,100+ client engagements in 60+ countries over 20+ years. It comes with practice questions that show the right answer. Each course ends with a final exam and a verifiable certificate that anyone can check online, plus a LinkedIn badge, and client administrators can show an auditor exactly who has been trained. You will be able to apply AI Risk Management Training (ISO/IEC 23894, NIST AI RMF and EU AI Act) at work and show your competence with confidence. Sterling Academy courses carry the same content as Sterling's classroom training, whose certificates are trusted and accepted by all major certification bodies, as evidence of training at client audits for over 20 years.
It is for AI risk and governance leads, risk managers, model risk and validation teams, compliance, privacy and legal staff, product and data science managers and internal auditors. It suits any organisation that builds, buys or uses AI systems and needs AI risk handled as part of enterprise risk management, whether or not it seeks ISO/IEC 42001 certification.
The course applies ISO/IEC 23894:2023 guidance on AI risk management, with ISO 31000:2018. It uses the NIST AI Risk Management Framework 1.0 and the Generative AI Profile, NIST AI 600-1. It explains the EU AI Act risk duties, Articles 9, 26 and 27, from the official text as amended by Regulation (EU) 2026/1744.
Practice exercises after key modules let you apply each topic to your own organisation and get feedback; they do not count toward the result. The course ends with a scenario-based multiple-choice final exam. Learners who score 60% or more receive a Sterling Academy Certificate of Competence, and one retake is included. Each certificate carries a unique number and a QR code for online checking.
The AI Risk and Impact Assessment course covers the risk and impact assessments that ISO/IEC 42001 requires for certification. This course shows how to run AI risk management as an ongoing programme: appetite, roles and three lines, ISO/IEC 23894, NIST, EU AI Act duties, suppliers, monitoring and board reporting, with or without certification.
The course is in preparation. Register on this page with “Notify me when it opens” and we will send you one e-mail when it opens for enrolment, with the price per learner and the details for team enrolment. The course page will then show the full module list and the video runtime.
Sources
Your certificate
You will be able to apply AI Risk Management Training (ISO/IEC 23894, NIST AI RMF and EU AI Act) at work and show your competence with confidence. Your certificate can be verified online in seconds, so certification body auditors, employers and clients can confirm your training instantly.
Reach the pass mark in the final exam and you receive a Sterling Academy Certificate of Competence, verifiable online by QR code, with a LinkedIn badge to share. Employers, clients and auditors can check it in seconds. Every learner who finishes the course receives a certificate: a Certificate of Completion when the retake does not reach the pass mark.
Sterling Academy courses carry the same content as Sterling's classroom training, whose certificates are trusted and accepted by all major certification bodies, as evidence of training at client audits for over 20 years.
Ask about course requirementsYour next step
Not sure which course?