AI-generated concept image: Hong Kong city backdrop and agent identity card; not an official credential, system interface, or actual payment service
AI can help people organize procurement lists and may further initiate procurement on behalf of enterprises. Moving one step further, the question extends from “whether the answer is accurate” to “who allowed it to do so.”
On November 2, the Hong Kong Monetary Authority announced the launch of the SAGE project to support the responsible development and application of agentic finance in Hong Kong. The project’s full name is Securing Agentic Finance with Governance and Enablement, and it advances application exploration together with governance arrangements.
According to the official announcement, potential scenarios include consumer-facing agentic shopping and everyday payments, enterprise-facing procurement and treasury management, and machine-to-machine usage-based payments for digital services. These are directions the project intends to explore and validate; the announcement does not mean that these services have already been launched across the board.
Alongside scenario exploration is research into AI agent identity and authorized permissions. The HKMA mentions “Know Your Agent” (KYA) measures and identity authentication mechanisms, while listing interoperability standards and legal and regulatory issues such as governance, liability, and consumer protection as part of its scope of work.
This gives “agent” a more concrete meaning than a chat interface. For the party receiving an instruction, knowing only that a request comes from a piece of software is not enough; it also needs to understand whom the software represents, what it is permitted to do, and whether the authorization is still valid.
Take enterprise procurement as a hypothetical scenario for understanding this issue: an assistant may be authorized to collect quotations and prepare orders but not authorized to make payments. Even if it has found a lower-priced supplier, payment authority cannot be inferred from the task goal alone. The boundaries between preparation, confirmation, and execution need to be clearly expressed in business processes.
If a task spans multiple services, this boundary must also be passed on. The procurement system recognizing this assistant does not automatically mean that the payment service also has the same authorization information. How both parties understand a single delegation and how they handle permission changes are practical issues that interoperability discussions may face. These are scenario analyses developed from the issues in the announcement, not technical solutions already published by SAGE.
The HKMA has established an Agentic Finance Expert Group with members from financial institutions, technology service providers, professional service firms, and academia to support the project’s work. This composition also shows that agentic finance involves more than model development: how business is executed, how users understand authorization, and who handles problems all require joint discussion among different stakeholders.
For ordinary users, whether an agent service is convenient in the future will not depend only on whether it saves a few taps on the screen. Users also need to understand what permissions they have handed over, know which steps require re-confirmation, and be able to find records that can be checked when a dispute arises.
What SAGE has currently established is a platform for advancing applications and discussing governance. What is worth watching next is how these issues become concrete validation scenarios and public arrangements, so that “handle it for me” becomes a delegation with a clear scope and an explainable process.