Which AI regime applies to you depends on where you operate and where you sell. DIFC Regulation 10 covers high-risk AI in the DIFC. The EU AI Act reaches any system placed on the European market, wherever the provider is based. NIST AI RMF anchors US enterprise and federal procurement. In the UK, obligations sit with existing sector regulators rather than a single act. Banks in the UAE also have CBUAE guidance to consider.
A firm headquartered in Dubai that sells into Europe and works with US enterprise clients can face three or four of these at once. The common response is a separate project for each: one consultant for ISO, another for the EU AI Act, a third for a client questionnaire. Each project builds its own policies, its own risk register and its own evidence. The firm pays for the same work several times and ends up with documents that do not agree with each other.
The frameworks ask for largely the same controls
Read side by side, these frameworks ask for the same core disciplines:
- Risk assessment: knowing which AI systems you run, what they decide and how badly they could go wrong.
- Data quality: evidence that the data behind each system is fit for its purpose.
- Transparency: people affected by a system can find out that AI is involved and how it is used.
- Human oversight: a named person can review, override or stop a system, especially for high-impact decisions such as credit.
- Security: the system is tested against misuse and attack before it reaches customers.
- Monitoring: performance and risk are tracked after go-live, not only at launch.
The wording, the thresholds and the documents each regime wants differ. The underlying work does not.
Start from ISO/IEC 42001
ISO/IEC 42001 is a management system standard for AI. It is the base every other framework attaches to, because it defines how an organisation governs AI rather than how one system behaves. Build the management system once, then layer the specific requirements of each regime on top as a framework pack.
In practice that means one set of unified controls (we use 24 across 12 governance domains), one document library and one evidence registry. Each control is pre-mapped to the clauses it satisfies in each framework. An auditor for any regime sees the same evidence, organised the way their framework expects.
The crosswalk does the translation
The key artefact is a certification crosswalk: a table that shows, control by control, how one piece of evidence satisfies each regime. It turns "are we compliant with the EU AI Act?" from a new project into a lookup. It also shows the gaps clearly, so effort goes only where a framework asks for something the others do not.
Frameworks change. Because controls are built once and mapped outward, a new regulation becomes an update to your crosswalk rather than a new project.
What 90 days looks like
- Weeks 1 to 2, scope. Build the AI inventory, including third-party models, run an applicability screen to confirm which frameworks apply, and name the governance roles.
- Weeks 3 to 8, implement. Stand up controls from tested templates. A live governance tracker with dashboard, risk register and evidence registry runs from week 3. Complete impact assessments for every in-scope system.
- Weeks 9 to 12, map and close. Map evidence to each target framework, close the remaining gaps, and run an internal audit and management review.
- Month 4 onwards, certify. Pursue certification with accredited bodies, reusing one evidence set.
Where certification is already expected, as it is for high-risk AI in the DIFC, starting from a ready-made implementation rather than a blank page saves months.
Governance that lets you move faster
Good governance is not a brake. When controls are built in before deployment, use cases reach production with fewer late surprises, and the board can approve the next one with confidence. The aim is one evidence base, owned by your team, that stands up to every auditor you face.
