What to know
- Project SAGE is an HKMA initiative for responsible exploration of agentic finance, not evidence that autonomous banking services are already deployed.
- Its four workstreams include use cases, agent identity and delegated authority, interoperability standards, and legal or regulatory questions.
- The announcement leaves practical questions about authorization, accountability and consumer safeguards for future work rather than resolving them.
Direct answer: Hong Kong’s Monetary Authority launched Project SAGE on 2 November as a program to explore agentic finance alongside governance work. Its announcement identifies four workstreams—use cases, agent identity and delegated authority, interoperability, and legal or regulatory frameworks. It does not announce a production banking agent or a new binding rule.
What the initiative actually covers
The HKMA describes potential business-to-consumer uses such as agentic commerce and everyday payments, business-to-business procurement or treasury tasks, and machine-to-machine usage billing. These are areas for exploration and validation with industry participants. The distinction matters: a list of possible use cases is not proof that a service is available to customers, has passed operational testing, or may act without human approval.
Identity and delegated authority are unusually concrete parts of the agenda. A financial system needs to distinguish the person or organization that grants permission, the software agent acting under it, and the exact action that permission covers. A label such as “trusted agent” cannot answer whether an instruction is still valid, what transaction limit applies, or who is responsible when a workflow produces an unexpected result.
Why the four workstreams belong together
Identity without bounded authority can authenticate an agent that still has excessive permissions. Authority without interoperable standards may work only inside one provider’s platform. Both can fail to protect customers if legal responsibility, dispute handling and consumer safeguards remain unclear. SAGE’s combination of these topics gives readers a useful evaluation lens: ask how an agent is identified, what it is allowed to do, how systems exchange that permission, and how a contested action is handled.
BijliCoin’s existing agent tool boundary definition explains the engineering distinction between a model proposing an operation and a connected tool being authorized to execute it. SAGE’s announcement operates at a broader industry and policy level, but the same separation helps make the initiative’s identity and delegation questions concrete.
What remains open
The HKMA says an expert group will support the project. The release does not specify a finalized technical standard, common identity credential, liability allocation, consumer redress process or timetable for production deployment. Nor does it establish that the proposed “Know Your Agent” concept is already a formal regulatory requirement. Those details will determine whether a future system can safely carry permission across institutions.
For banks and vendors following the work, useful questions include whether delegated permissions can be narrowed by amount, recipient and purpose; how revocation propagates; what evidence is retained; and whether customers can see and challenge actions. These are evaluation questions, not claims about controls already delivered by SAGE. The announcement is best read as a research and coordination agenda whose value will depend on the safeguards and implementation evidence that follow.
AI-assisted editorial preparation and an AI-generated illustration were used. Claims are attributed to the linked sources; corrections may be submitted through BijliCoin’s contact page.
Sources
- HKMA: Project SAGE and Agentic Finance Expert Group (2 November 2026)
Editorial note: AI-assisted editorial preparation and an AI-generated illustration were used. Claims are attributed to the linked sources; corrections may be submitted through BijliCoin’s contact page.
Sources & context
See our editorial policy or report a correction.
Related guides and definitions
- Build an agent evaluation scorecard around accepted outcomes
A practical method for comparing agent changes without confusing fluent output, tool success or benchmark rankings with useful work.
- Choose a bounded workflow before giving an agent open-ended control
A practical design guide for deciding what an AI system may choose, what code should enforce and where a person remains responsible.
- What is an agent tool boundary?
An agent tool boundary is the enforced separation between what a model may propose and what a connected tool is authorized to do.
Join the conversation.
Loading reader discussion…
Sign in to like this article or add a comment. Comments are reviewed before appearing publicly.
Open the sign-in link in this same browser. Your display name is public; your email is private. Reader sign-in does not subscribe you to the newsletter. See our privacy notice.
Signed in as



