The joint circular of 5 March 2026 from the HKMA, the Securities and Futures Commission, the Insurance Authority and the MPFA expands the HKMA's GenA.I. Sandbox into the GenA.I. Sandbox++, covering banking, securities and capital markets, asset and wealth management, insurance, MPF and stored value facilities. Participants get complimentary GPU computing resources at Cyberport's A.I. Supercomputing Centre, targeted supervisory guidance from their own regulator and technical support from Cyberport. Applications were open until 30 June 2026; for banks the channel was the HKMA's Survey Tool platform. The first cohort, announced on 27 August 2026, selected 36 use cases focused on agentic AI.
OFFICIAL TEXT: brdr.hkma.gov.hk ↗ · IN FORCE · HKMA
| Document | HKMA GenA.I. Sandbox++ joint circular — Joint Circular on the Expansion of Generative Artificial Intelligence Sandbox (HKMA, SFC, Insurance Authority and MPFA, 5 March 2026) |
| Issued by | Hong Kong Monetary Authority — Hong Kong's banking regulator and de facto central bank (circulars to Authorized Institutions under the Banking Ordinance, Cap. 155) |
| Type | Circular |
| Status | In force |
| Published | Mar 5, 2026 |
| Applies to | Authorized Institutions under the Banking Ordinance (Cap. 155), licensed corporations under the Securities and Futures Ordinance, authorized insurers and licensed insurance broker companies, MPF approved trustees and principal intermediaries, and stored value facility licensees. Participation is voluntary. |
| Official source | brdr.hkma.gov.hk ↗ |
| Use cases | Generative & agentic AI · Fraud detection · Customer-facing chatbots · Credit scoring & underwriting · AI governance (general) |
What are the key points of HKMA GenA.I. Sandbox++ joint circular?
- Scope: an expansion of the GenA.I. Sandbox launched in 2024 for Authorized Institutions to the sectors of the four financial regulators (HKMA, SFC, IA and MPFA), in partnership with Cyberport.
- Support: complimentary GPU computing at Cyberport's A.I. Supercomputing Centre, targeted supervisory guidance from each regulator and technical support from Cyberport; GenA.I. Sandbox Collaboratory workshops to turn problem statements into use cases.
- Application window: open until 30 June 2026; Authorized Institutions apply through the HKMA's Survey Tool platform (GenAI_sandbox@hkma.gov.hk for enquiries).
- Focus areas in the Annex: risk management (for example creditworthiness evaluation from unstructured data), anti-fraud measures (including deepfake scams and forged documents) and customer experience (personalised chatbots).
- Safety expectation: all use cases should incorporate A.I. safety and risk management components, including bias detection and mitigation, explainable A.I. and frameworks for A.I. output monitoring and evaluation; "A.I. vs. A.I." strategies are encouraged.
- Data: participants are encouraged to apply data minimisation, masking or tokenisation, and may use public, anonymised or synthetic data; a secure data transfer and access mechanism is provided.
- Evaluation factors: level of innovation, complexity of the solutions, expected contribution to the industry and adherence to the principle of fair use.
- First cohort: announced 27 August 2026 with 36 use cases from nearly 100 proposals, 30 financial institutions and 27 technology partners, focused on agentic AI.
What did HKMA GenA.I. Sandbox++ joint circular change for banks?
It extended the HKMA's bank-only sandbox to the whole financial sector under four regulators and moved its themes toward AI safety and agentic AI, creating a single cross-regulator entry point for supervised AI pilots in Hong Kong.
How does the HKMA GenA.I. Sandbox++ work and what does it require of participants?
The GenA.I. Sandbox++ is a joint initiative of the HKMA, SFC, Insurance Authority and MPFA, with Cyberport, that lets regulated firms across banking, securities, insurance, MPF and stored value facilities pilot AI and GenAI solutions with complimentary GPU computing, targeted supervisory guidance from their own regulator and Cyberport technical support. Applications ran until 30 June 2026, with banks applying through the HKMA's Survey Tool platform. Use cases should centre on risk management, anti-fraud and customer experience and incorporate A.I. safety components such as bias detection, explainable AI and output monitoring; participants must allocate sufficient internal resources and provide regular progress updates to their regulator. The first cohort, announced on 27 August 2026, has 36 use cases focused on agentic AI.
| Rule | Authority | What it requires | Applies |
|---|---|---|---|
| Joint circular, eligibility — Who may apply | HKMA | Authorized Institutions, licensed corporations, authorized insurers and licensed insurance broker companies, MPF approved trustees and principal intermediaries, and stored value facility licensees. | Applications closed 30 June 2026 |
| Annex, application procedures — Banks apply to the HKMA | HKMA | Authorized Institutions apply through the HKMA's Survey Tool platform, giving high-level design, applicable models, risk assessments and information about technology partners. | Closed 30 June 2026 |
| Annex, objectives — A.I. safety and risk management | HKMA | All use cases should incorporate A.I. safety and risk management components, including bias detection and mitigation, explainable A.I. and frameworks for A.I. output monitoring and evaluation. | For each use case |
| Annex, general principles — Data minimisation and security | HKMA | Participants are encouraged to apply masking or tokenisation, may use public, anonymised or synthetic data, and are expected to implement adequate data security controls on data used in the Sandbox. | For each use case |
| Annex, expectations on participants — Resources and reporting | HKMA | Allocate sufficient internal resources and provide regular updates to the relevant regulator on project progress. | During participation |
| Press release of 27 August 2026 — First cohort ↗ | HKMA | The first cohort of 36 use cases focuses on agentic AI, including testing of end-to-end processes such as onboarding, payments, insurance claims and customer interactions, and AI providing dynamic oversight of other AI. | Announced 27 August 2026 |
Sandbox++ builds on two bank cohorts of the GenA.I. Sandbox and reuses its Collaboratory workshop model. The circular cites strong industry response and the exploration of "A.I. vs. A.I." strategies, including AI automated quality checks on AI-generated outputs.
It is a supervised-pilot programme; the HKMA's AI principles of 1 November 2019 and its 19 August 2024 GenAI consumer-protection circular continue to apply to what banks launch after piloting.
WHAT THIS MEANS IN PRACTICE
- Treat Sandbox participation as a way to obtain supervisory feedback on agentic or GenAI use cases before customer-facing rollout.
- Build bias detection, explainability and output monitoring into pilot design, since the circular expects them in every use case.
- Apply data minimisation and secure-transfer controls to pilot data, using masked, tokenised, anonymised or synthetic data where possible.
- Plan resource allocation and periodic updates to the HKMA for the duration of the pilot.
What is the HKMA GenA.I. Sandbox++?
An expansion of the HKMA's GenA.I. Sandbox, launched by the HKMA, SFC, Insurance Authority and MPFA with Cyberport on 5 March 2026. It adds securities, insurance, MPF and stored value facility firms to banks and provides GPU compute, supervisory guidance and technical support.
Can banks apply to the GenA.I. Sandbox++?
Authorized Institutions under the Banking Ordinance were eligible and applied through the HKMA's Survey Tool platform; applications were open until 30 June 2026. The first cohort was announced on 27 August 2026.
What kinds of use cases does Sandbox++ look for?
Primarily risk management, anti-fraud measures and customer experience, with every use case expected to include A.I. safety and risk management components such as bias detection, explainability and output monitoring. The first cohort focuses on agentic AI, including onboarding, payments and customer interactions.
Does the GenA.I. Sandbox++ relax regulatory requirements?
The circular describes supervisory guidance and technical support rather than relaxations. Participants are expected to provide regular updates to their regulators on project progress.
| Date | Document | Status |
|---|---|---|
| Sep 20, 2024 | HKMA GenA.I. Sandbox circular — Generative Artificial Intelligence Sandbox (HKMA circular inviting applications to the GenA.I. Sandbox, 20 September 2024) | In force |
| Sep 9, 2024 | HKMA circular on AI for monitoring of suspicious activities — Use of Artificial Intelligence for Monitoring of Suspicious Activities (HKMA circular, 9 September 2024) | In force |
| Aug 19, 2024 | HKMA GenAI consumer protection circular — Consumer Protection in respect of Use of Generative Artificial Intelligence (HKMA circular, 19 August 2024) | In force |
| Nov 5, 2019 | HKMA BDAI consumer protection principles — Consumer Protection in respect of Use of Big Data Analytics and Artificial Intelligence by Authorized Institutions (HKMA circular, 5 November 2019) | In force |
| Nov 1, 2019 | HKMA High-level Principles on Artificial Intelligence — High-level Principles on Artificial Intelligence (HKMA circular to all Authorized Institutions, 1 November 2019) | In force |
| Sep 30, 2026 | Bailey: Frontier AI and the Question of Governance (Sep 2026) — Frontier AI and the Question of Governance — Governor Andrew Bailey | Final |
Follow every document these regulators publish
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning