Regulation · 19 November 2026

AI governance framework in the UAE and GCC, mapped

Cover card for a Triway Technologies insight on a light background. Tag: Regulation. Title: AI governance framework in the UAE and GCC, mapped. Triway Technologies logo in the corner.

There is no single AI governance framework for the UAE or the GCC; each organisation's framework is assembled from the instruments that reach it. In the UAE those are the national AI charter, Regulation 10 of the DIFC Data Protection Regulations for entities in the Dubai International Financial Centre, and the Central Bank of the UAE's guidance note on AI for licensed financial institutions; in Saudi Arabia, the AI Ethics Principles of the Saudi Data and Artificial Intelligence Authority (SDAIA); and, for groups with European customers, ISO/IEC 42001 and the EU AI Act. Which of them apply depends on where an organisation is licensed, the sector it operates in and whose personal data its systems touch, and every one of them asks for evidence that only a running system can produce.

Triway Technologies builds and runs AI systems for organisations that answer to these instruments, so this note maps them the way an implementer reads them: what each one is, who it reaches, and which controls in a live system produce the evidence it asks for. The gap we meet most often is not a missing policy. It is a policy that describes intentions and an estate in which nothing enforces them.

Which AI governance instruments apply in the UAE and Saudi Arabia?

Five instruments, and the data protection laws beneath them, do most of the work.

  • UAE Charter for the Development and Use of Artificial Intelligence. The national charter, issued on 10 June 2024, sets out twelve priorities, among them safety, algorithmic bias, data privacy, transparency, human oversight, and governance and accountability. It serves the goals of the UAE Strategy for Artificial Intelligence and sets direction rather than prescribing controls.

  • DIFC Data Protection Regulations, Regulation 10. Enacted on 1 September 2023, Regulation 10 covers personal data processed through autonomous and semi-autonomous systems in the Dubai International Financial Centre. It provides for notice to individuals when such systems are in use, processing limited to human-defined or human-approved purposes, a register of each system's use cases and processing, available on request, an Autonomous Systems Officer where the processing is high risk, and audit and certification requirements set by the Commissioner of Data Protection.

  • CBUAE guidance note on AI and machine learning (11 February 2026). The Central Bank of the UAE's guidance note on consumer protection and the responsible adoption and use of AI and machine learning applies to the financial institutions it licenses, insurers included. It sets expectations on documented governance proportionate to the institution, fairness testing, explainability to consumers, three models of human oversight, an inventory of AI models with purpose and risk rating, disclosures in Arabic and English, and accountability that stays with the institution when AI comes from a third party.

  • SDAIA AI Ethics Principles (version 1.0, September 2023). SDAIA's AI Ethics Principles apply to entities designing, developing, deploying and using AI in the Kingdom. They set seven principles, specify controls for each across four lifecycle phases from planning to monitoring, and sort systems into four risk levels, from little or no risk to unacceptable.

  • International references. ISO/IEC 42001:2023 specifies the requirements for an AI management system. The EU AI Act reaches organisations whose AI systems are placed on the market or used in the European Union.

Underneath all of them, wherever personal data is involved, sit Federal Decree-Law No. 45 of 2021 on the protection of personal data and, in the financial free zones, the DIFC and ADGM data protection regimes.

What do the UAE and GCC instruments ask for in common?

Read side by side, the instruments ask the same five questions in different words.

  • What AI do you run, and for what? An inventory of each system, its purpose, the data it touches and its risk rating. The CBUAE note describes a model inventory with name, purpose and risk rating; Regulation 10 describes a register of use cases and processing.

  • Who decides? A defined level of human involvement per use case. The CBUAE note names three models: human-in-the-loop, human-on-the-loop and human-out-of-the-loop, the last reserved for low-risk, non-material processes.

  • Can you explain it? The logic behind an outcome, in terms the affected person understands, and in Arabic and English where the CBUAE note applies.

  • Is it fair, and does it still work? Testing for bias and drift before release and after every change, not once at go-live. The CBUAE note expects bias testing at least annually and after material changes.

  • Who is accountable? A named owner inside the organisation, including when the model or the platform comes from a vendor.

What changed in AI governance in 2026?

Three changes matter for Gulf organisations this year.

  • The CBUAE note arrived. Issued on 11 February 2026, it gives UAE banks and insurers a central bank reference point for AI governance, built on the bank's existing model management standards.

  • The DIFC is revising Regulation 10. In June 2026 the DIFC published Consultation Paper No. 3 of 2026, proposing amendments that add safety as a core design concept, set out privacy-by-design measures, clarify certification obligations, detail the Autonomous Systems Officer's duties and introduce a new Regulation 11 under which the Commissioner of Data Protection can recognise accreditation and certification frameworks. The consultation closed on 18 July 2026; at the time of writing we have not seen the amended regulations enacted.

  • The EU AI Act timetable moved. The Act became generally applicable on 2 August 2026, and the AI Omnibus, in force since 27 July 2026, moves the high-risk rules for Annex III use cases to 2 December 2027.

Why is AI governance a systems problem rather than a policy?

A policy document answers none of those questions on its own; each answer has to be produced by the system itself, on demand, as evidence a supervisor can inspect.

In practice the controls look like this: an inventory generated from what is actually deployed rather than maintained by hand; guardrails set per use case, covering what each system may do, which data reaches it and which outputs need a person; an evaluation harness that reruns a fixed test set on every model or prompt change and records the result; decision logs that can reconstruct why an output was produced; and disclosures written once, in Arabic and English, and served wherever the system meets a customer.

The levels of human oversight deserve particular care. The three CBUAE models map directly onto how we build agents that carry work across systems: the agent drafts and a person approves (in the loop); the agent acts and a person monitors and can intervene (on the loop); the agent acts alone within hard limits on low-risk work (out of the loop). Choosing the level is a design decision per task, recorded where an auditor can see it and revisited as evidence builds up. We wrote about that dial in what an AI agent is allowed to do.

Where do we start with AI governance?

Start with the inventory, because every other control depends on it, and because most organisations find more AI in use than they expected: copilots switched on inside productivity suites, vendor features enabled by default, and tools teams adopted on their own. Then take the highest-risk system on the list and build its controls end to end: guardrails, evaluation, logging, disclosure and a named owner. One governed system teaches an organisation more than a framework document, and it becomes the template for the rest.

For banks and insurers in the UAE, the CBUAE note now gives that work a clear reference point. For government entities, the national charter sets the direction, and the same controls produce the evidence.

That is how our ninety-day method treats governance: inventory and risk classification during the diagnostic, controls built into the first system rather than added after it, and a governance playbook your team inherits at handover. In the two-week diagnostic we list the AI you already run, including the copilots and vendor features nobody registered, and rate each system against the instruments that reach you.

This note describes the published rules as we understand them at 19 November 2026. It is not legal or compliance advice.

Related on triway.ai

Begin here

Start with a 30-minute call

Free. We ask what you run and what you want to change, and tell you whether the two-week diagnostic is worth your money. If it is, you leave it with a plan worth keeping, whoever you choose to build with.

Not ready to talk? See where AI would pay off first: twelve questions, three minutes.