What AI rules apply to banks in Qatar?

The Board and senior management remain accountable for the outcomes and decisions of the Entity's AI Systems, including systems that make decisions on behalf of the Entity. The Board approves the level of AI exposure tolerated in the overall risk framework, decides whether existing governance structures are fit for purpose, and ensures adequate human resourcing of all AI functions. Senior management must include one or more members with the knowledge to understand and manage technology risks.

The function responsible for overseeing AI must utilize or create appropriate committees to assess AI use cases prior to implementation. An Entity should establish a function overseeing AI or delegate that responsibility to an existing function.

High-Risk AI is defined at §2 as systems risk-assessed as having potential to cause significant negative impact to the Entity's operations or the financial system, judged by severity, intensity, probability, duration, and ability to affect individuals or limit access to financial resources or essential services. Notwithstanding the Entity's own determination under §9.6, it must classify a system as high-risk where there is potential harm to natural persons in three areas: interactions determining consumer access to financial services, internal decisions materially affecting employees, and processing of sensitive personal information.

For any new high-risk AI System, an Entity must submit the full results of the system's training, validation and testing to the QCB for approval. A User must additionally submit the Provider's instructions for use and the Provider's technical and training results, plus the results of its own use testing, its data sources and its human oversight plan.

An Entity must receive official QCB approval prior to signing any high-risk AI purchase, licensing or outsourcing agreement. Before granting approval the QCB may direct the system into a sandbox environment for further evaluation. Separately, approval is required before launching a new AI System as a Provider or materially modifying an existing one.

A high-risk system with no human oversight of decision execution requires prior QCB approval before launch, and the Entity is expected to notify the QCB as soon as such a system is actively being considered or developed. Even where a fully autonomous system is viewed as low or no risk, the Entity must provide very detailed information to support its use for QCB approval.

A fully autonomous high-risk system must have built-in guard rails or specific limits the AI cannot override, reviewed on a frequent set schedule or in response to external volatility spikes, with in-built limits linked to warning levels or auto-close routines. The Supervisor must be able to close the system down where outputs or surrounding data seem aberrant.

Any AI System must have a human oversight protocol. The User assigns oversight to a Supervisor with the competence, training and authority to operate or oversee the system, equipped to understand its capacities and limits, interpret its output, decide not to use or to disregard, override or reverse the output, and interrupt the system through a stop button or similar procedure. Systems must carry built-in operational constraints that the system itself cannot override.

An Entity must develop and maintain an updated Register of all its AI System arrangements, recording per system the high-risk classification, whether the Entity is User or Provider, a category reflecting functional use, and the human oversight protocol used.

The full Register goes to the QCB annually and on request. The Entity must separately disclose the criteria it uses to decide whether a contract for a specific system is high-risk, and a high-level risk and impact assessment.

For every high-risk purchased, licensed or outsourced system the register must rate the AI's substitutability as easy, difficult or impossible and identify an alternate service Provider where applicable, alongside governing law, contract start and renewal dates, and the provider's country of registration and LEI.

An Entity must be able to manage a system across its life cycle either directly or by contractual provision from the Provider, who is expected by contract to supply all data from development, testing, pre and post-market introduction, regular performance testing and system development. The Entity must make available to the QCB the full range of data the Provider is contractually obliged to provide to it.

Board approval is required to outsource any AI-related function, and must be documented. Service agreements must include the right to terminate for non-compliance or when the QCB so directs, exit clauses ensuring data is wiped on termination, and a QCB right to audit the provider's accounts. The Entity remains responsible as principal for all acts of its outsourcing providers.

The Entity must ensure the model is checked for systematic bias by testing it on different demographic groups, to observe whether any group is being systematically advantaged or disadvantaged. Training, validation and testing data sets must be subject to documented design choices and data governance practices covering collection, preparation, assumptions, prior suitability assessment, bias examination and identified data gaps.

An Entity must develop an AI Trust, Risk and Security Management program including toolsets for content anomaly detection, data protection from Providers and third parties, and third-party system security, plus an acceptable use policy and processes for assessing privacy, fairness and bias. The program must ensure AI Providers adhere to it.

Examine an AI model's attack surface prior to deployment and address findings; protect against integrity attacks, against query attacks leading to manipulation or theft, and against prompt injections leading to data poisoning or drift; deploy a data loss prevention tool and content anomaly detection tools.

An Entity must support monitoring systems and plans for high-risk systems, including accessing Provider information, and must ensure Provider compliance with evaluating conformance to predicted model outcomes across the life cycle. Records should be auditable: audit logs and traceability of decisions and outcomes, design documentation, robust model versioning including code, and archiving of the original data sets used to develop, re-train or calibrate models.

Customers must be notified when interacting with an AI System, given plain-language disclosure, told on request the data types and factors influencing the decision, and told how a decision affects them and whether it is reversible. A mechanism must exist to request review of decisions made with no human intervention, offering a two-choice process: supply data to alter the system and resubmit, or request human review by a qualified decision-maker. Intellectual property need not be disclosed, and fraud and identity-theft models are excluded from disclosure.

An Entity seeking exemption from any requirement must request it from the QCB, supported by a documented business case and linked to a specific requirement. All exemptions are subject to QCB approval, must be recorded in an exemptions Register with an expiration date by which the exemption will be mitigated or resolved, and the Entity remains accountable for the risks arising from them.

Artificial Intelligence Guideline (2024)

Qatar Central Bank

Titled a guideline, drafted as instructions with a commencement clause, and operating as a pre-authorization regime. It distinguishes must from should deliberately: the approval gates, register, bias testing and security controls are must; customer consent, the opt-out and most documentation duties are should. This is why this entry is labeled binding where the UAE entry is not: the QCB instrument carries a commencement clause and mandatory drafting, and the CBUAE Guidance Note carries neither.

22 pages, 24 articles across six parts, read in full. The PDF carries no text layer: its text is outlined vectors. Pages were therefore rendered to images with pypdfium2 and read directly rather than extracted. §1 states: the instructions set forth herein are titled the Artificial Intelligence Guideline for 2024 and shall enter into force as of 04/09/2024.

Law No. (13) of 2016 on Personal Data Privacy Protection

State of Qatar

The data protection floor beneath the AI Guideline. §15.15 permits processing personal data strictly as necessary for bias work, subject to this law, with pseudonymization or encryption where feasible.

Cited at §15.15 as the law governing processing of personal data by Providers of high-risk systems for bias monitoring, detection and correction. The statute itself was not separately reviewed for this entry.

Sector-Specific Security Regulation

Qatar Central Bank

AI security obligations sit on top of the existing sectoral security regime rather than replacing it.

Referenced at §17.1, which requires compliance in all deployment of AI solutions. Not separately reviewed for this entry.

Technology Risks Circular (January 2018)

Qatar Central Bank

Part of the stack §24 wires the AI Guideline into, in the same way the CBUAE note routes into the Model Management Standards.

Named at §24.1.3 as a secondary regulation an Entity must comply with alongside the AI Guideline. Not separately reviewed for this entry.

Cloud Computing Regulation (2024)

Qatar Central Bank

Applies to most production AI deployments in practice, since almost all of them are cloud-hosted.

Named at §24.1.4 as applying where an entity leverages cloud-based AI solutions. Not separately reviewed for this entry.

FinTech and Digital Transformation Strategy (2025)

Qatar Central Bank

Context rather than obligation. Signals AI supervision is being institutionalized rather than left as a single publication.

Reported as expanding the 2024 guideline into a wider strategy making AI oversight a core pillar. Not reviewed in primary form.