Legal Accountability for AI-Driven Insurance Decisions in the UAE

time 4 min 18 sec

Artificial Intelligence (“AI”) can support decisions across the insurance value chain, including underwriting and risk assessment. For a policyholder whose claim has been rejected, what matters is whether the decision is justified. Automation can make that justification harder to uncover, particularly where an insurer relies on data and models the customer cannot see.

For insurers operating under the UAE’s federal framework, the answer requires consideration of insurance regulation, the insurance contracts, civil liability and data protection. The Central Bank of the UAE (CBUAE) also addresses governance, fairness and human oversight in its Guidance Note on Consumer Protection and the Responsible Adoption and Use of Artificial Intelligence and Machine Learning by Licensed Financial Institutions in the U.A.E., issued in February 2026 (the “Guidance Note”). It applies to Licensed Financial Institutions within the scope of Federal Decree-Law No. 6 of 2025 Regarding the Central Bank, Regulation of Financial Institutions and Activities, and Insurance Business (the “Federal Law”). The Guidance Note sits alongside the applicable laws and regulations; it does not replace them.

Who is responsible for the AI?

Responsibility for AI governance rests with the institution using the system. The Guidance Note calls for oversight by the company’s Board of Directors and Senior Management, documented arrangements, reporting and scrutiny by the functions responsible for checking risk and compliance. For an insurer, this means keeping enough control over the model to understand its use, assess its outputs and challenge a result where necessary.

The Federal Law regulates the activity regardless of the technology used. An insurer’s obligations therefore do not change simply because a decision is automated. If it rejects a customer’s complaint or claim, in whole or in part, it must give written reasons. Those reasons should explain the relevant facts and policy terms, not only state that an automated assessment led to the rejection.

Human oversight on automated decisions

Human oversight is an important element emphasized in the Guidance Note. The Guidance Note takes an approach based on the risk and the likely effect on the customer. Activities without direct human involvement should be limited to those that are low-risk and non-material, with appropriate safeguards. The more significant the decision, the more important it is that someone can examine and challenge the result.

Customers should have a clear way to ask for an explanation or human review of a significant decision and to challenge inaccurate information. The Guidance Note also calls for alternatives for customers who do not wish to undergo AI decision-making. Disclosures in Arabic and English should help customers understand the process and how to use those options.

Data protection law provides a separate framework applicable alongside the Guidance Note. Federal Decree-Law No. 45 of 2021 Concerning the Protection of Personal Data (the “PDPL”) states that individuals have rights to object to automated decisions, subject to exceptions, and to request human review. Therefore, insurers should check which rights apply to the particular decision rather than assume that every automated process is prohibited and ensure that it is done in compliance with the regulatory framework.

Reliable data and responsible use

An AI-driven decision is only as reliable as the information it is fed with. Incorrect or outdated data may affect the assessment of the AI even where the model operates as intended. The Guidance Note emphasises accurate, relevant and current information, identifiable sources and appropriate security. Regular testing for bias, including when a model is introduced or materially changed, can help identify patterns that may lead to unfair customer outcomes.

However, insurers must also consider whether they are permitted to use the data for the intended purpose. The PDPL requires lawful processing, limits on the data collected and the correction of inaccurate personal data. Health data is also subject to the relevant health authorities’ regulatory framework. Similarly, insurers operating in financial free zones have their own data-protection laws, such as the Dubai International Financial Centre and the Abu Dhabi Global Market.

For health insurers, this includes restrictions on handling UAE health data overseas. Permission to use or disclose information for insurance purposes should not be treated as permission to use it for unrelated AI training.

Who is responsible when an AI decision causes harm?

An automated flag by AI does not automatically justify refusing a claim. For example, a system may identify a claim as potentially fraudulent. That may be a reason to investigate, but it does not by itself show that the loss falls outside the policy’s cover. The insurer still needs to examine the evidence, apply the policy terms and decide whether it has grounds to withhold payment. The distinction matters: a model’s assessment of risk is not the same as a conclusion about the cover.

Using an external provider adds another question: who contributed to the harm? The answer depends on the facts and the court’s assessment. Liability may be shared, but equal or joint and several liability, where a party may be required to answer for the whole loss, is not automatic. Any right the insurer has to recover from its provider is also separate from the policyholder’s claim against the insurer. Contractual arrangements cannot exclude or reduce liability arising from harmful acts.

An automated flag by AI does not automatically justify refusing a claim. For example, a system may identify a claim as potentially fraudulent. That may be a reason to investigate, but it does not by itself show that the loss falls outside the policy’s cover.

Third-party AI providers: managing outsourced decision-making

The use of third-party AI providers does not displace the insurer’s responsibility for oversight. The Guidance Note addresses provider selection, due diligence, pre-deployment assessment and independent annual cybersecurity reviews. Information and audit rights support continuing supervision of the outsourced service and reduce reliance on the provider’s own assurances.

The checks should continue throughout the arrangement. Insurers should monitor material changes, test updates and retain the ability to stop using the system through human intervention. Where the PDPL applies, the arrangement should also protect personal data and set out the instructions under which the provider may process it. These controls matter particularly where the insurer cannot directly see how the provider’s model works.

Records matter when a decision is disputed. The insurer should be able to identify the information used, the model version, the reasons recorded and any human review or intervention. These records can help explain what happened and whether the decision followed the policy. UAE evidence and electronic-transactions laws recognise electronic records subject to the applicable conditions, but the records do not prove the insurer’s position simply because they exist.

Conclusion

AI can support insurance decisions, but it does not replace the rules governing the insurer or the policy. The Guidance Note sets out expectations for responsible use, while the applicable laws determine obligations and remedies. Insurers should therefore check their review processes, provider arrangements and decision records before a dispute arises. Customers need a clear route to question an outcome, and the insurer needs enough information and control to examine it and, where appropriate, input the right information.