The roster of firms selected for Hong Kong’s first GenA.I. Sandbox++ cohort – AXA, FWD Life, HSBC Life, BOC Group Life, and Manulife among them – signals that agentic AI in insurance is no longer a future consideration. It is an active deployment question, and regulators are writing the governance rules in real time. The Generative Artificial Intelligence Sandbox++ (GenA.I. Sandbox++), announced August 27, 2026, is a joint initiative of the Hong Kong Monetary Authority (HKMA), the Securities and Futures Commission (SFC), the Insurance Authority (IA), and the Mandatory Provident Fund Schemes Authority (MPFA), in collaboration with Hong Kong Cyberport Management Company Limited. Thirty-six use cases were selected from nearly 100 proposals, involving 30 financial institutions and 27 technology partners. For every licensed broker and intermediary in Hong Kong who is not among those 30 institutions, the key question is not what the sandbox participants are testing. It is what compliance obligations will look like for the rest of the market once testing concludes.
The published use case list provides a clearer picture of the areas being tested. Proposed applications include an AI knowledge assistant for medical claims, intelligent fraud prevention for digitally altered medical documentation, a multi-agent system for fraud detection and behavioural risk monitoring, and agentic payment systems. Technology partners include Google, IBM, Tencent Cloud, PwC and MediConCen, a Hong Kong-based medical claims technology company. The range of use cases highlights the programme’s focus on applying AI across financial services workflows, including insurance.
Regulators said the tests cover end-to-end processes such as customer onboarding, payments, insurance claims, and customer interactions. The pilots will also explore multi-agent applications involving interactions between AI systems, bringing a more complex form of AI deployment into insurance use cases. Similar multi-agent and agentic AI concepts have also been explored through the HKMA’s earlier GenAI sandbox work.
In July 2025, the International Association of Insurance Supervisors (IAIS) published its Application Paper on the supervision of AI, reinforcing the importance of the Insurance Core Principles (ICPs) and outlining how existing expectations around governance and conduct remain relevant to insurers using AI. The paper does not establish new standards or expectations but provides guidance on applying existing supervisory principles to AI, including governance, accountability, human oversight, transparency, fairness, and redress.
For brokers and intermediaries, the practical implication is direct: AI deployment does not create new obligations, but it does require firms to demonstrate compliance with existing conduct standards under materially more complex conditions – including when AI is independently processing claims, detecting fraud, or communicating with policyholders. SFC CEO Julia Leung stated: “The financial industry’s exploration of agentic AI marks a pivotal step forward in developing a future-ready Hong Kong market, where greater system autonomy is underpinned by sound governance and clear accountability.”
The sandbox announcement sits within a broader pattern of regulatory escalation that brokers cannot treat in isolation. The IA received 1,173 complaints throughout 2025, broadly consistent with pre-pandemic levels but occurring in the context of continued market growth. Complaints relating to conduct and representation of information, which together made up 45% of the total, were largely associated with intermediaries’ pre-contract sales practices and post-sale servicing.
The authority has also warned that enforcement actions against brokers are not one-off measures, with Alan Wu, acting head of conduct supervision, stating: “These actions are not isolated initiatives and we will not end from there.” The IA has placed conditions on the license renewals of two unnamed brokers after finding failures in referral activity controls. For intermediaries, the trajectory is a single regulatory direction of travel: tighter governance expectations across distribution, remuneration, and AI-driven workflows, enforced simultaneously.
The IA indicated in August 2025 that updated AI guidelines will be issued in 2026 to provide regulatory clarity on the use of AI and facilitate responsible adoption without creating unnecessary regulatory frictions or hurdles. As of publication, those guidelines have not yet been released – making this a live gap in the compliance calendar. The IA has also been explicit about the intended reach of sandbox learnings beyond the participating institutions. IA chairman Stephen Yiu said at the AI Cohort Symposium in June 2026 that the IA encourages core participants to share practical experiences with brokers and smaller market participants, so that SMEs can adopt AI in a proportionate way and the market can advance together. IA CEO Clement Cheung said: “The GenA.I. Sandbox++ has inspired new ideas, generated practical insights, and deepened cross-sector collaboration. The IA will strive to reinforce the position of Hong Kong as an international innovative and technology centre under the National 15th Five-Year Plan through progressive expansion of the AI Cohort Programme.”
The HKMA’s first banking-focused GenA.I. Sandbox cohort reported that institutions cut preparation work for Suspicious Transaction Reports by 30% to 80%, reduced memo processing from a full day to minutes, and shortened the production time for high-quality outputs by 60%, with 86% of GenA.I. outputs rated favourably by users. Insurance claims carry a materially different risk profile – coverage disputes, fair treatment obligations, intermediary conduct standards – but the efficiency case for agentic AI is already established in adjacent regulated workflows.
Technical trials are scheduled to commence later in 2026, with participants expected to be onboarded to the Cyberport Artificial Intelligence Supercomputing Centre platform. Regulators said insights from the trials will inform their ongoing engagement with the industry, while proposed deployments will remain subject to applicable regulatory, governance, privacy and risk-management requirements. Participation in the sandbox does not itself constitute regulatory approval for commercial deployment.
Brokers who are waiting for the IA’s forthcoming AI guidelines before engaging with governance questions around agentic AI are likely to find, when those guidelines arrive, that the accountability expectations they set have been shaped by a testing process in which they had no voice.
