Hong Kong Monetary Authority Launches Project SAGE to Govern AI Agents in Finance

Hong Kong's central bank launched the most comprehensive government-led governance framework for AI agents in financial services on November 2, convening 22 institutions including Visa and UnionPay to build rules for autonomous payment and treasury operations. Project SAGE — Securing Agentic Finance with Governance and Enablement — is the Hong Kong Monetary Authority's direct answer to a question every major regulator is still avoiding: who is accountable when an AI agent makes a financial decision autonomously?

The HKMA unveiled Project SAGE at Hong Kong FinTech Week on November 2. The initiative forms a 22-member Expert Group that will draft frameworks covering four problem areas that no existing financial regulation adequately addresses: AI agent identity verification, interoperability between competing agent systems, consumer protection in autonomous transactions, and legal liability assignment when an AI agent causes financial harm.

Why this is different from prior AI finance guidance

Previous regulatory output on AI in finance — from the Financial Stability Board, the Bank for International Settlements, and the Basel Committee — addressed AI risk in general terms. The guidance covered model risk, explainability, and bias in AI-assisted decisions. It was designed for a world in which AI advises humans, who then execute.

Agentic AI changes the model entirely. An AI agent in treasury management does not recommend a position for a trader to approve; it executes the trade. An AI agent in payments does not flag a suspicious transaction for review; it blocks or releases the payment. The accountability chain that existing financial regulation assumes — human decision, traceable to a named individual — breaks down when the decision-maker is an autonomous system.

SAGE is the first initiative specifically designed to address that break.

The Expert Group and what it will build

The 22-member Expert Group includes Visa Hong Kong and UnionPay International — payment infrastructure companies whose networks will carry transactions initiated by AI agents. Their participation is not ceremonial: the interoperability standards SAGE develops will need to integrate with existing payment rails, meaning Visa and UnionPay need to be in the room while the frameworks are written.

The four framework areas SAGE will tackle:

• AI agent identity verification: How a financial institution confirms that an AI agent acting on behalf of a customer or counterparty is authorized to do so, and that it is the agent it claims to be — a problem analogous to digital identity but applied to software acting as a principal
• Interoperability: How AI agents from different institutions, built on different models, communicate and transact without creating new systemic fragmentation
• Consumer protection: What disclosures, consent requirements, and redress mechanisms apply when an AI agent, not a human, makes a financial decision affecting a consumer
• Legal liability: When an autonomous AI agent causes financial harm — through an erroneous trade, an unauthorized payment, or a misconfigured treasury operation — who is legally responsible and under what standard

No implementation timeline has been announced. The Expert Group's findings will feed into HKMA guidance, not yet into enacted regulation.

What this signals for the industry globally

The HKMA is the first major central bank to create a dedicated agentic AI governance body. That precedent matters because financial regulation tends to propagate across jurisdictions: when a tier-one regulator like the HKMA, MAS, or the Fed establishes a framework, other regulators use it as a starting template rather than building from scratch.

For fintech companies operating in Asia-Pacific, SAGE signals that a compliance obligation is forming well before any legislation is enacted — and that the time to engage with the Expert Group's process is now, not after frameworks are finalized.

For payment networks globally, the Visa and UnionPay participation is a signal that interoperability standards for AI agent transactions are entering serious development. Any payment company not involved in that conversation risks being excluded from the infrastructure layer that governs how AI agents move money.

What to watch

When the HKMA's Expert Group publishes its first findings report — timing not yet announced — examine the proposed liability framework most closely. The question of whether liability for an AI agent's action falls on the deploying institution, the model provider, or the agent's operator is the most commercially consequential decision SAGE will make. Watch also for whether BIS or the FSB cite SAGE frameworks in any upcoming guidance documents, which would confirm that Hong Kong's approach is becoming the global baseline.