US Banks Draft Liability Rules for AI Agent Payments
Lead Bank proposed a 'waterfall' framework and startup Primitive drafted the Agentic Financial Services Protocol, as digitaltoday.co.kr reported.
US banks, payment firms and technology companies are working to set standards that assign liability for financial transactions carried out by AI agents, according to digitaltoday.co.kr. US bank Lead Bank proposed a "waterfall" framework that determines liability based on the cause of an incident, while fintech startup Primitive put forward a draft standard called the Agentic Financial Services Protocol (AFSP) to verify AI identity and user authorization and to trace transactions for auditing and regulation.
Under Lead Bank's proposal, consumers are not liable when an AI agent exceeds its delegated authority. Banks are liable when they fail to control that authority, developers are liable for AI model defects or insufficient safeguards, and sellers are liable for intentionally inducing an AI agent to exceed its authority.
Regulatory reform seen as the bigger task
Jackie Reses, CEO of Lead Bank, said reforming related banking and payment regulations is a bigger task than technically controlling liability.
On September 29, Federal Reserve Board governor Christopher Waller said at a Sibos event that lack of trust is the biggest obstacle to the spread of agentic commerce, and that existing consumer protection rules and payment network rules need to be supplemented.
Consulting firm Bain projected that annual US spending on AI agent commerce will reach $500 billion by 2030.
Quick answers
What is the Agentic Financial Services Protocol (AFSP)?
It is a draft standard proposed by fintech startup Primitive to verify AI identity and user authorization and to trace transactions for auditing and regulation.
How does Lead Bank's 'waterfall' framework assign liability?
Consumers are not liable when AI exceeds delegated authority, banks are liable when they fail to control authority, developers are liable for AI model defects or insufficient safeguards, and sellers are liable for intentionally inducing AI to exceed its authority.
What did the Federal Reserve say about agentic commerce?
On September 29, Federal Reserve Board governor Christopher Waller said at a Sibos event that lack of trust is the biggest obstacle to the spread of agentic commerce, and that existing consumer protection rules and payment network rules need to be supplemented.