HK Insurance AI Underwriting: Rules and Deployment

The regulatory framework and current deployment

Hong Kong’s insurance sector is the area where AI exploration has been most concrete among the financial regulated industries. The Insurance Authority (IA) launched several structural facilitation programmes in 2025 and 2026, but current regulation has not yet drawn specific boundaries for AI automation in writing, and the industry is finding its way forward.

The IA’s Facilitation Mechanisms

In August 2025 the IA launched the AI Cohort Programme, bringing together insurance leaders and technology service providers with a focus on underwriting, claims, fraud detection, customer service and sales support. Seven major insurers joined: AIA, AXA, FWD Group, Prudential, HSBC Life, YF Life Insurance and China Taiping Insurance (Hong Kong).

In March 2026 the HKMA, SFC, IA and MPFA launched GenA.I. Sandbox++ together with Cyberport, providing a regulated environment to test cross-sector AI applications, with insurance underwriting and claims processing within its scope. The first cohort selected 36 use cases from nearly 100 proposals, and AXA, FWD Life, HSBC Life, BOC Group Life and Manulife were all among those selected.

The Current State of the Regulatory Framework

Hong Kong’s Insurance Ordinance (Cap. 41) currently has no express provision drawing the specific boundary between AI automation and human decision-making. The generally accepted industry principle is that AI assists with processing and analysis, while licensed professionals are responsible for final approval. Actuarial judgement, decisions on high-value claims, and underwriting steps that require medical verification all retain human review in industry practice.

Data Processing Compliance Considerations

Insurance underwriting involves large volumes of personal health and medical data, which is highly sensitive personal data. The “AI: Model Personal Data Protection Framework” published by the PCPD in June 2024 gives a guidance framework for handling such data in AI systems, emphasising compliance with the collection purpose limitation principle under the Personal Data (Privacy) Ordinance (PDPO).

In the mixed Chinese-English forms common in Hong Kong insurance (with stamps and handwritten fields), the accuracy of AI document recognition is limited by what current OCR technology can do with non-standard layouts. System design should include flagging of low-confidence recognition results and a human re-check mechanism, rather than defaulting to the assumption that every extracted result is accurate.

For the difficulty of mixed-language documents, see Half Chinese, Half English: Where AI Systems Fail; for methods and costs of turning unstructured documents into structured data, see AI Document Processing Automation.

Anti-Discrimination Compliance

Hong Kong’s Sex Discrimination Ordinance, Disability Discrimination Ordinance, Race Discrimination Ordinance and Family Status Discrimination Ordinance all apply to insurance underwriting decisions. If an AI underwriting model uses proxy variables related to protected characteristics (such as postal district or occupation category), it may amount in effect to indirect discrimination, even when the model design does not explicitly use a protected characteristic as an input.

Bias testing before model deployment is a compliance requirement and goes beyond a purely technical matter. In its November 2024 AI circular the SFC already required licensed corporations in the securities industry to conduct bias testing on AI models. The insurance industry has no identical mandatory requirement, but the same principle applies.

The same indirect discrimination principle applies to hiring; see AI Recruitment Screening in Hong Kong: Legal Limits, Bias Risk and Practical Approaches.

Summary

The current state of AI in Hong Kong insurance is that regulators are actively advancing pilots and sandbox programmes and large industry participants have entered the practice stage, while the specific boundaries at the level of regulation still depend on industry self-discipline and the continuing evolution of regulator guidance. Until regulation becomes explicit, retaining human review points and complying with data processing requirements are the two dimensions that most need to be secured first when deploying an AI underwriting system.

Further reading: Hong Kong Accounting Firm AI: What to Automate