US Banks Restrict AI Agents Moving Funds while HKMA Explores Identity and Authorization Frameworks

As AI agents gain capabilities to issue financial transactions for clients, financial institutions and regulators in the US and Hong Kong are navigating risks and governance. In the United States, major institutions including JPMorgan Chase, Charles Schwab, Citi, and Bank of America have restricted external AI agents from directly moving funds or executing payments, citing rogue agent risks and unresolved liability frameworks, generally limiting tools to read-only queries. Conversely, the Hong Kong Monetary Authority (HKMA) announced Project SAGE (Securing Agentic Finance with Governance and Enablement) on November 2, forming an expert group with 22 industry members—including HSBC, Standard Chartered, Bank of China (Hong Kong), Visa, and AWS—alongside an academic representative. Concurrently, HKMA published a discussion paper on agentic AI governance in banking. HKMA preliminarily suggests treating external agents as instruction-submitting conduits, separating identity verification (KYA) from authorization checks via deterministic controls and transaction caps, while maintaining instant revocation capabilities. For banks' internally deployed agents, proposed measures include intent-anchored accountability, emergency kill switches, and central registration. Industry consultation runs through February 1, 2027.

Key takeaways

• Major US institutions including JPMorgan, Schwab, and Citi restricted external AI agents from moving funds due to unresolved liability rules.
• The HKMA launched Project SAGE with a 22-member expert group to explore agent identity (KYA), interoperability standards, and legal accountability.
• HKMA's discussion paper suggests treating external agents as instruction conduits, decoupling identity from authorization under strict limits.
• For banks' proprietary AI agents, proposed controls include intent-anchored accountability, emergency kill switches, and central registry reporting.