OceanAltOceanAlt
Agent Payments2026-11-18Event 2026-11-175 min read

Fed and HKMA Speak on the Same Day: Two Compliance Tracks for Agent Payments Are Converging

Fed Governor Barr and Hong Kong officials on Nov 16 separately pointed to the same set of issues: institutional accountability for agents, and verification of authorization when agents act on behalf of clients.

OOceanAlt EditorialSource ↗

Federal Reserve Governor Michael Barr and Hong Kong regulatory officials each spoke on agentic AI on Monday, Nov 16, pointing to the same set of unresolved issues: how institutions are accountable for the AI agents they deploy, and how institutions determine whether an agent acting on behalf of a client is truly authorized to move funds or make financial decisions. According to PYMNTS, Barr said at an international financial AI conference in Milan that large banks have encountered cases where agents attempted to execute operations in retail customer accounts, but the banks could not verify their authenticity, the customer's consent, or consistency with instructions; he did not name the banks or provide the number of incidents.

Barr laid out three requirements: strict identity authentication, explicit verification of consent for specific operations, and technical controls that limit agent capabilities. These three correspond to the judgment made before settlement—who is initiating, whether the customer has authorized it, and whether the operation falls within the authorized scope. Notably, the problem he identified goes beyond banks' internal technology: customers may bring autonomous systems into financial relationships before institutions have reliable means of identification. In other words, the object of authorization verification is not only the agents institutions deploy themselves, but also agents coming in from the client side.

He also warned of another risk path: agents could accelerate deposit outflows during periods of stress. Systems monitoring interest rate spreads or instability signals would automatically transfer funds, potentially compressing a bank run into minutes and leaving conventional liquidity management and regulatory intervention insufficient time to respond. This is not abstract model risk but a problem of speed—a structural mismatch between the execution tempo of machine-to-machine payments and traditional liquidity management tools measured in days or hours.

On the same day in Hong Kong, discussion at the Bloomberg Global Regulatory Forum took a different path: institutional accountability and emerging regulatory arrangements. According to the South China Morning Post, participants discussed risks including market volatility and cybersecurity. Financial Secretary Paul Chan's remarks were cited as a footnote to this approach: "Algorithms have no license to lose and no reputation to protect. Institutions have both." This pins responsibility on the institutions deploying agents and their management.

Chan mentioned the HKMA's Project SAGE, launched earlier this month, which studies agent identity and authorization delegation, interoperability standards, and legal issues including liability allocation, governance, and consumer protection, with potential applications covering consumer payments, corporate procurement and treasury management, and machine-to-machine payments. On another front, Securities and Futures Commission CEO Julia Leung acknowledged in a separate speech on Nov 2 that manual review of every intermediate step is unrealistic; according to the report, the SFC plans to advance related arrangements.

Placing the two jurisdictions' remarks side by side, it is clear that regulatory approaches are converging along two dimensions. The first is accountability: when an agent errs or exceeds its authority, responsibility falls on licensed institutions and people, not algorithms. The second is authorization: institutions need to be able to confirm an agent's identity, what it is permitted to do, and when humans can intervene. The former is a governance issue, the latter a technical and procedural one, and in practice the two intersect at the same point—before funds move out.

This is where the signal of an emerging compliance framework for agent payments lies. Know Your Agent (KYA) answers "who is this agent, and whom does it represent"; mandate and per-transaction limits, daily cumulative limits, and payee whitelists answer "what can it do, and to what extent"; pre-settlement interception places these two layers of judgment before funds are transferred, rather than discovering anomalies during post-hoc reconciliation. What Barr described as "unable to verify authenticity, consent, and consistency" essentially means at least one of these three layers is missing. Hong Kong's inclusion of agent identity and authorization delegation in Project SAGE's research scope shows that regulators recognize that internal governance clauses alone cannot constrain cross-institutional, cross-border agent behavior; interoperable standards are also needed.

The implications differ for different stakeholders. For banks and brokerages, agent identification and authorization verification will move from "nice to have" to part of account opening and onboarding processes; since manual review of every step is unrealistic, control points must be shifted forward to the rules and limits level. For payment service providers and wallets, onboarding client-facing agents requires a credential system that is attributable and clearly non-repudiable; otherwise, when disputes arise, it will be impossible to reconstruct "who authorized what." For regulators, Barr's bank-run warning and Hong Kong's accountability framework point to the same policy toolkit: capability limits plus real-time visibility, rather than ex-post penalties.

One question that the remarks from both jurisdictions did not directly answer: when client-owned agents and institution-deployed agents interact within the same account system, how are liability boundaries drawn. Project SAGE's inclusion of liability allocation and consumer protection in its research scope indicates this remains an open item. Before standards are settled, the realistic choice for institutions is to make authorization verification and limit controls default pre-settlement steps, rather than waiting for regulatory details to be issued before adding them.


Source: PYMNTS · https://www.pymnts.com/news/artificial-intelligence/2026/regulators-converge-on-twin-compliance-challenges-from-ai-agents/

Provenance & status

Byline
OceanAlt Editorial
First published
2026-11-18
Last updated
2026-11-18
Content type
Original compilation
Source material
View original ↗

Cite this piece

OceanAlt Editorial (2026). "Fed and HKMA Speak on the Same Day: Two Compliance Tracks for Agent Payments Are Converging". OceanAlt. https://oceanalt.com/en/articles/deep-auto-mwkxuvva-ouaf (accessed 2026-11-21)

This piece follows our editorial and fact-checking standards. Found an error? tell us. Once verified, the correction will be published right here.

TRY IT · FREE, NO SIGNUP

Paste a payee address before you pay and see whether it's on a sanctions list, through a mixer, or tagged for fraud.

This judgement can sit inside your own product

One line of code; it touches neither your CSS nor your JS. The same pre-settlement judgement can appear in your articles, on your wallet's confirmation screen, or as an endpoint your agent calls before it pays.

The widget collects no reader identity. Integrating does not mean OceanAlt endorses your product, or any address on your page.