Hong Kong’s financial regulators are moving generative AI testing beyond chatbots and content generation toward systems that can actually execute workflows. Dyna.Ai has been selected as a technology partner in the first cohort of GenA.I. Sandbox++, a joint initiative involving the Hong Kong Monetary Authority (HKMA), Securities and Futures Commission (SFC), Insurance Authority (IA), Mandatory Provident Fund Schemes Authority (MPFA) and Cyberport. The selection puts Dyna.Ai among technology providers testing how agentic AI can perform complex financial tasks while remaining subject to governance and supervisory controls.
The financial industry’s AI experiment is entering a more consequential phase.
Early generative AI deployments largely focused on assistants that could summarize documents, answer questions, generate content or interact with customers. The next generation is expected to do more: retrieve information, make decisions within defined boundaries, interact with enterprise systems and execute multi-step workflows.
That transition is now being tested under regulatory supervision in Hong Kong.
Dyna.Ai has been selected for the first cohort of GenA.I. Sandbox++, a program jointly launched by the HKMA, SFC, IA and MPFA in collaboration with Cyberport. The initiative expands Hong Kong’s earlier banking-focused GenAI sandbox into securities, insurance and pension-related financial services while putting agentic AI at the center of its trials.
Nearly 100 proposals were submitted for the program. Regulators selected 36 use cases involving 30 financial institutions and 27 technology partners, with technical trials scheduled to begin later this year. The selected projects will examine areas including customer onboarding, payments, insurance claims and customer interactions.
For Dyna.Ai, the selection is also a second consecutive appearance in a Hong Kong regulatory AI sandbox, following its participation in an earlier HKMA initiative.
From AI assistants to systems that take action
The difference between generative AI and agentic AI is important for financial institutions.
A conventional AI assistant might summarize a customer’s financial documents or answer an employee’s question. An agentic system can potentially go several steps further: gather information from multiple systems, determine the next action, execute it and monitor the result.
That additional autonomy creates both an opportunity and a new category of risk.
A bank could theoretically use an AI agent to coordinate onboarding, retrieve documentation, check requirements and escalate exceptions. An insurer could use an agent to gather information associated with a claim and initiate parts of the claims workflow.
But when AI moves from generating information to taking action, traditional controls around authorization, auditability and human oversight become more important.
Hong Kong’s regulators are explicitly addressing that issue. The Sandbox++ program says selected trials will examine how AI can responsibly operate with greater autonomy, while also exploring an “AI vs. AI” model in which one AI system can provide dynamic oversight of another.
That makes the sandbox more than a technology showcase. It is effectively a controlled environment for testing how autonomous financial software should behave under regulatory supervision.
Governance becomes part of the product
Dyna.Ai says its participation will focus on its agentic AI capabilities and governance frameworks, with the objective of demonstrating how financial institutions can delegate workflow execution while retaining oversight and compliance controls.
That emphasis mirrors a broader problem emerging across enterprise AI.
McKinsey’s 2026 AI Trust Maturity Survey found that responsible-AI maturity is improving, but strategy, governance and agentic-AI controls continue to lag technical development. Only about one-third of organizations reached higher maturity levels in those areas.
Financial services faces an even higher bar.
Banks, insurers, securities firms and pension providers operate within regulatory frameworks where decisions can affect money, credit, investments, insurance coverage and customer rights. An AI system that makes an incorrect recommendation is one problem; an autonomous system that executes the wrong transaction is considerably more serious.
That is why the industry’s move toward agentic AI is likely to depend on governance architecture as much as model performance.
Hong Kong is broadening the AI test bed
The expansion from the original GenAI Sandbox is significant.
The HKMA launched its initial sandbox to give banks a controlled environment for experimenting with AI use cases. The second cohort expanded the practical applications to areas such as risk management, fraud prevention and customer experience.
Sandbox++ broadens the regulatory perimeter further.
The four financial regulators are now testing AI across banking, securities and capital markets, insurance, mandatory provident funds and stored-value facilities. Participants also receive access to GPU computing resources through Cyberport’s Artificial Intelligence Supercomputing Centre alongside targeted supervisory feedback.
That creates an unusually broad test environment for agentic financial AI.
It also positions Hong Kong as a proving ground for technologies that could eventually be deployed across Asia’s financial markets.
Competition is moving beyond chatbots
Dyna.Ai is entering a market where large technology companies and specialist financial-AI vendors are pursuing similar ambitions.
Google, IBM and other technology providers are participating in the Sandbox++ cohort, alongside financial institutions including Bank of China (Hong Kong) and Manulife, according to Dyna.Ai’s announcement.
Globally, Microsoft, Google, AWS and Salesforce are building agent frameworks and enterprise AI platforms that can connect models with business applications and workflows.
The differentiation for financial AI vendors is increasingly shifting from the underlying large language model to the surrounding system: integrations, domain knowledge, permissions, observability, workflow orchestration and governance.
That is particularly relevant in regulated industries, where enterprises may not want an autonomous agent operating as an unrestricted general-purpose software user.
What financial institutions should watch
For banks and other financial institutions evaluating agentic AI, the Sandbox++ approach highlights several requirements.
First, agents need clearly defined permissions. Institutions must be able to determine which actions an agent can perform autonomously and which require human approval.
Second, organizations need auditability. Every significant action should be traceable to the underlying data, instruction, policy and decision path.
Third, AI systems need continuous monitoring. An agent that behaves correctly during testing may encounter unexpected conditions once connected to live financial workflows.
McKinsey’s research on banking AI similarly points to the importance of reusable data foundations, security controls, agent autonomy limits, monitoring and audit mechanisms as banks move toward more autonomous systems.
The broader market is moving in the same direction.
Financial institutions have already spent years testing generative AI for productivity and customer service. The next challenge is moving from AI that assists employees to AI that participates in the execution of financial processes.
Hong Kong’s Sandbox++ provides a regulatory setting for testing that transition.
For Dyna.Ai, the selection offers an opportunity to demonstrate its technology under precisely those conditions. For the financial sector, the more important outcome may be the evidence regulators and institutions collect about where agentic AI can safely operate—and where human intervention must remain mandatory.
The future of financial AI is unlikely to be defined simply by how autonomous an agent can become.
It will be defined by how much autonomy institutions can safely govern.
Market Landscape
The financial AI market is moving through three overlapping stages: generative AI assistants, workflow automation and agentic AI.
The first wave focused largely on employee productivity, search, document processing and customer-facing chatbots. The emerging wave connects AI models to enterprise systems so they can execute multi-step processes.
That creates a fundamentally different risk profile.
Hong Kong’s regulators are responding by expanding the sandbox beyond banking and conventional generative-AI applications. Sandbox++ covers banking, securities and capital markets, insurance, MPF and stored-value facilities, while focusing on use cases such as onboarding, payments, insurance claims and customer interactions.
The competitive landscape includes:
- Microsoft — Copilot and Azure AI agent capabilities across enterprise applications.
- Google — Gemini and Google Cloud’s enterprise AI and agent ecosystem.
- AWS — Amazon Bedrock and agent-development infrastructure.
- IBM — watsonx and enterprise AI governance.
- Salesforce — Agentforce for autonomous customer and business workflows.
- Specialist fintech AI vendors such as Dyna.Ai, which focus on domain-specific financial workflows.
The differentiator is increasingly not simply the model. Financial institutions need an orchestration layer capable of connecting AI to proprietary data, applications and regulated processes while enforcing permissions and maintaining audit trails.
That is why governance is emerging as a competitive feature rather than a compliance afterthought.
Top Insights
- Dyna.Ai has entered Hong Kong’s first GenA.I. Sandbox++ cohort, putting its agentic AI platform into regulator-supervised testing across complex financial workflows and autonomous operations.
- Sandbox++ selected 36 use cases from nearly 100 proposals, involving 30 financial institutions and 27 technology partners across Hong Kong’s broader financial ecosystem.
- Hong Kong regulators are shifting from generative AI toward agentic systems, testing autonomous workflows in onboarding, payments, insurance claims and customer interactions.
- AI governance is becoming a core financial technology requirement, as autonomous agents introduce greater operational, compliance and accountability risks than conventional AI assistants.
- Financial institutions are moving toward controlled AI autonomy, requiring permissions, monitoring, human escalation and audit mechanisms before agents can execute production workflows.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
