AI in UAE Financial Services: A Practical Legal Guide
By Ankita Dhawan and Anika Prakash
AI is now in everyday use across financial services in the UAE. Banks, insurers, investment managers, payment providers, fintechs and virtual asset service providers use it for fraud detection, onboarding and KYC, customer service, credit decisions and underwriting, transaction monitoring, compliance, investment tools and risk management. Notably, using AI does not reduce, transfer or remove any of the financial service provider's existing legal responsibilities.
No single UAE law governs AI in financial services. Instead, AI now runs through the whole value chain: how a customer is onboarded and assessed, how a product is priced, advised on and sold, how transactions are monitored, and how complaints are handled. Each of those stages is already regulated, so wherever AI enters, the existing rules follow it. That makes compliance exposure as broad as its use of AI. It covers financial regulation and licensing, consumer protection and market conduct, data protection, cybersecurity and operational resilience, outsourcing and third-party risk, AML/CFT, and a growing body of AI-specific guidance. For an onshore bank, an insurer, a payment provider, a fintech or a VASP, the crucial question is how all of these laws interact with each other.
This guide takes each of these in turn, both for financial institutions deploying AI and for the technology companies building it for them.
// Do you need a licence, and if yes, from which regulator?
Which regulatory framework applies to AI tools being built for financial services depends on what the business does and where it is licensed.
Onshore banks, finance companies, insurers and payment providers are supervised by the Central Bank of the UAE (CBUAE) under Federal Decree-Law No. 6 of 2025 (the New CBUAE Law). It came into force on 16 September 2025, and its one-year reconciliation period, which the CBUAE Board may extend, ran to September 2026.
Onshore capital markets activity is regulated by the Capital Market Authority (CMA). The CMA replaced the Securities and Commodities Authority on 1 January 2026 under Federal Decree-Law No. 32 of 2025 and Federal Decree-Law No. 33 of 2025.
Firms in the financial free zones are regulated by the Dubai Financial Services Authority (DFSA) in the DIFC and the Financial Services Regulatory Authority (FSRA) in ADGM. Both zones have their own rulebooks and their own data protection laws. The New CBUAE Law excludes the financial free zones, although free zone firms that target UAE retail customers onshore may still be exposed under the federal framework.
Virtual asset businesses are licensed by VARA in the Emirate of Dubai outside the DIFC, by the DFSA or FSRA inside the free zones, and by the CMA elsewhere onshore under its April 2026 Virtual Assets Framework. Payment tokens fall under the CBUAE’s Payment Token Services Regulation
For AI companies offering services in the financial sector, the key question is whether they fall inside the regulatory perimeter. The New CBUAE Law is technology-neutral. Article 62 requires anyone who carries on, offers, issues or facilitates a licensed financial activity, by whatever medium or technology, to be licensed by the CBUAE. It expressly covers platforms, applications, protocols and technology infrastructure that facilitate or enable financial services such as payments, credit, deposits, exchange, remittances and investments. So the use of AI does not take an activity outside the regime.
However, equally, providing technology that supports a regulated financial activity does not automatically mean the technology provider itself requires a financial services licence. The CBUAE has clarified in its FAQs that Article 62 does not create new licensed activities. It has also clarified that services provided exclusively for the benefit of a licensed financial institution are not treated as carrying on, offering, issuing or facilitating a licensed activity. The test for an onshore AI vendor is therefore whether it is supplying a tool to a licensed firm or providing the financial service itself, for example by making credit decisions, giving robo-advice or orchestrating payments directly for customers. A vendor that markets its product as the service should assume it is inside the perimeter.
In the DIFC, the DFSA’s FinTech guidance indicates that a business may not need DFSA authorisation where it is a technology provider or is partnering with regulated firms. Once a provider carries on a regulated activity itself, such as advising on investments or arranging deals, it will need a licence.
// Are AI tools covered by governance frameworks?
Even where an AI provider does not need its own licence, it does not escape regulation. The licensed financial institution that buys its product remains fully accountable for how that product behaves, and UAE regulators expect that accountability to sit with the board and senior management. In practice, the institution’s governance obligations become the vendor’s product requirements. An AI tool that a bank or insurer cannot document, validate, monitor or switch off is a tool it cannot lawfully use.
Licensing is therefore only one layer of regulation. Alongside it, UAE regulators have built a set of governance principles and guidance that shape how licensed institutions must adopt and oversee AI, and, through them, what they will require from the AI providers they work with. Two instruments matter most.
The first is the Guidelines for Financial Institutions Adopting Enabling Technologies, which the CBUAE, the SCA (now the CMA), the DFSA and the FSRA jointly issued in November 2021, well before the current wave of AI adoption. They apply to all institutions licensed by those regulators that use, or plan to use, AI and other enabling technologies. For AI, they call for:
a documented governance framework covering the full model lifecycle;
continuing accountability of the board and senior management for AI outcomes;
models that are reliable, transparent and explainable;
independent validation before deployment, periodic review, and the ability to suspend a model promptly;
audit logs, traceability and versioning;
plain-language disclosure to customers, and a way for them to request review of AI decisions.
The Guidelines are not legally binding, but they serve as the benchmark for assessing AI governance.
In February 2026, the CBUAE went further with its Guidance Note on Consumer Protection and Responsible Adoption and Use of AI and Machine Learning. It applies to all CBUAE-licensed financial institutions, including banks, finance companies and insurers. The Guidance Note is also not legally binding, but firms should expect it to form part of supervisory dialogue and regulatory assessments. It expects firms to:
adopt a documented AI governance framework proportionate to their size, nature and complexity;
place accountability for AI systems and their outcomes on the board and senior management;
keep an inventory of all AI models and systems they develop or deploy, including third-party models;
risk-rate each AI system.
Under the Guidance Note, financial institutions must not employ AI models that they have no control over. Where “black-box” models are used, the institution must have enough contractual access to documentation and model logic to meet its own transparency obligations. Therefore, commercially speaking, to be procurable by a regulated client, an AI product should come with:
clear documentation of how the model works, what data it uses and its known limitations;
validation and testing evidence that the client can rely on before deployment;
audit logs, traceability and version control;
outputs that can be explained to the client’s customers and its regulator;
the information the client needs for its AI inventory and risk-rating;
the ability for the client to suspend or switch off the system promptly.
// How does data governance across the value chain work?
Data laws apply at every stage of the AI value chain. They cover the customer data an AI system processes once it is live, and also the data used to train it. Which regime applies depends on location:
Onshore: the federal Personal Data Protection Law applies, although it excludes banking and credit data that is already governed by its own legislation.
ADGM: the ADGM Data Protection Regulations 2021 apply.
DIFC: the DIFC Data Protection Law No. 5 of 2020 applies, together with Regulation 10 of the DIFC Data Protection Regulations, which governs personal data processed through autonomous and semi-autonomous systems such as AI.
Regulation 10 applies to the “deployer”, which is broadly accountable for compliance, and to the “operator”, the provider that develops or runs the system on the deployer’s behalf.
Systems must be designed in line with its principles from the outset.
For high-risk processing, the system must be certified, and the deployer or operator must appoint an Autonomous Systems Officer.
The CBUAE Guidance Note also expects training data to be accurate, relevant and representative of the customers the model will serve, and documented so the model can be audited.
For AI providers, this means knowing where your training data comes from and not reusing client data to train your models without express agreement. In the DIFC, it also means building Regulation 10 into the system itself. For financial institutions, it means controlling what data goes to the vendor, where it is stored, whether it can be used for training, and what happens to it when the contract ends.
// How is Consumer Protection ensured?
Consumer protection law does not change because a decision is made by a model. Onshore, the CBUAE’s Consumer Protection Regulation and its Standards apply to every licensed financial institution. Their rules on disclosure and transparency, market and business conduct, protection of consumer data, and complaints therefore cover AI as much as they cover staff. In practice, this means:
a chatbot that gives wrong product information is a disclosure failure;
an AI pricing or credit model that treats customers unfairly is a conduct failure;
a complaint about an AI decision must be handled like any other complaint.
The CBUAE’s Guidance Note adds AI-specific protections on top: for “high-impact decisions” such as loan applications or insurance claims, customers should be told when AI is used, be able to request an explanation or human review, and be offered an alternative to an AI decision.
In the free zones, the DFSA’s and FSRA’s conduct of business rules play the same role.
For AI providers, this means your product will be judged against the client’s conduct obligations, so its outputs must be accurate, fair and explainable to a customer. For financial institutions, it means AI-driven interactions need the same oversight, testing and complaints handling as human ones.
// Can a licensed financial servies provider outsource the AI but keep control?
Buying AI from a third party is usually outsourcing. The licensed firm stays fully responsible to its regulator, so it passes those responsibilities down to the AI provider through the contract.
Institutions licensed by the CBUAE, the CMA, VARA, the DFSA and the FSRA are each subject to their own outsourcing and third-party risk requirements, such as the CBUAE’s Outsourcing Regulation for Banks. Their detail differs, but they share common expectations:
due diligence before appointing a provider;
ongoing oversight of the provider’s performance;
regulator access to information;
business continuity and exit planning;
timely reporting of incidents, including cyber incidents.
In practice, outsourcing contracts with AI providers can therefore include provisions such as:
audit, information and regulator access rights;
access to model documentation and logic;
restrictions on data use, including on model training;
controls on subcontracting;
incident and breach notification within set timelines;
notice of material model changes;
performance and accuracy commitments;
the right to suspend or switch off the system;
exit and transition support;
allocation of liability where the model gets it wrong.
// Does AI change AML/CFT obligations?
The answer depends on who is building the AI.
If you are a licensed institution building your own AI tools, your AML/CFT obligations do not change. Under Federal Decree-Law No. 10 of 2025, you remain responsible for customer due diligence, ongoing monitoring and reporting suspicious activity, whether that work is done by people or by a model. AI can make transaction monitoring, sanctions screening and customer risk scoring far more effective. Even so, any thresholds or tuning that reduce alerts must be documented, validated and defensible. When a transaction goes unreported, “the model did not flag it” is not a defence.
If you are a third party building AML/CFT solutions for licensed institutions, the law does not impose these obligations on you directly. Your clients carry them, and they will pass the requirements down to you through due diligence and contract. In practice, this means your tool must be:
explainable and auditable;
tested against each client’s risk profile;
transparent about what it does not detect;
supported by clear version control and change notification.
A licensed institution cannot rely on a monitoring tool it cannot validate, so a tool that fails these requirements will not get past procurement.
// Conclusion
AI is reshaping UAE financial services, but it does not change who is accountable. Licensing, governance guidance, data protection, consumer protection, outsourcing and AML/CFT rules all continue to apply wherever AI enters the value chain.
If you are building AI inside a licensed institution, every one of those obligations already sits with you. AI is a new way of meeting them, not an exception to them. Govern, test and monitor your models as you would any other critical function, with clear authority to change, suspend or retire them.
If you are building AI for licensed institutions, you may not need a licence yourself, but your clients’ obligations will reach you through procurement and contract.
Building AI for UAE financial services? Consilium Advisors helps AI developers and financial institutions map their regulatory perimeter, design governable products, and structure vendor arrangements that regulators and clients will accept.


