Six banks set AI agent payment principles but leave liability open
Six banks have published principles for AI agent payments. Who pays when an agent is tricked is still open, and insurance cover may not fit.
Analysis | By CoverWire, created using AI
Six banks, including NatWest, have agreed principles for payments that AI agents make on a customer's behalf. They have not said who carries the loss when an agent is tricked into paying the wrong party. CoverWire's analysis is that the gap matters to crime, cyber and financial lines underwriters, because existing reimbursement rules and policy wordings were written for people, not software.
Key points
- NatWest, ASB, Bank of America, Capital One, CBA and ING published shared principles for "agentic commerce" on 22 September.
- The banks describe the principles as a foundation for further discussion, not a finished standard, and invite others to help develop a blueprint for applying them.
- Bank of America said liability management is among the areas that will need thoughtful approaches.
- CoverWire's analysis: it is unverified whether UK scam reimbursement rules, or crime and cyber wordings, respond to a payment an agent was tricked into making.
What the banks have said
According to the NatWest Group release, the six banks set out principles in five areas: transparency, safety, privacy and data, choice, and interoperability. The banks call them a foundation for further discussion and invite others across payments to help develop a blueprint for applying them in practice.
In the same release, Bank of America said building confidence will require thoughtful approaches to identity, authorisation, fraud prevention, liability management and customer protection. CoverWire reads that as a sign the liability question is not yet settled.
Where existing rules might apply
The UK already has a regime for one kind of payment fraud. Under the Payment Systems Regulator's authorised push payment scam rules, victims who are tricked into sending money to a fraudster by Faster Payments or CHAPS between UK accounts can be reimbursed. The sending and receiving firms split the cost 50:50.
Those rules are framed around a person being deceived. Whether they cover a payment that a customer's AI agent was manipulated into making is unverified. CoverWire has found no UK case or regulatory statement on the point.
Where cover might apply
This section is CoverWire's interpretation, and the wordings have not been checked. If an agent pays a fraudster, a business might look to three places. Crime policies may cover social engineering, but may require that an employee was deceived. Cyber policies may cover funds transfer fraud after a system is compromised, which may not fit an agent that worked as designed but was fed false instructions. Professional indemnity may respond if a firm's own agent caused a client's loss.
The common thread is that each route was written around either a person being fooled or a system being broken into. An agent that follows bad instructions is neither. The idea that businesses will accept, tolerate or refuse agents, and that narrow access and specific approvals will decide trust, was raised by Nate B. Jones in a post on websites blocking AI agents.
The other explanation
The gap may be smaller than it looks. If banks require customers to approve each agent payment, the customer is still the one authorising it, and the existing rules and wordings may apply as they do now. On that reading, liability follows the human approval, and nothing new is needed.
The answer depends on how much freedom agents are given. Narrow, approved payments look like today's payments. Broad, standing permissions do not.
Any blueprint the banks and others develop may show which way they lean. No date has been given for one. Underwriters may also want to know how insureds that let agents spend are limiting what those agents can pay, and to whom.
Related coverage: Anthropic warns insurers may lack AI catastrophe capacity.
Sources
- NatWest Group, global banks collaborate on principles for trusted agentic commerce, 22 September 2026
- Payment Systems Regulator, APP scams
- Idea credit: Nate B. Jones, "Websites Are Blocking AI Agents", 31 October 2026
CoverWire is a news and information service. Nothing here is investment advice.
Free briefing
The CoverWire briefing
Get the day's UK insurance news by email at 08:30 every weekday. It's free, and you can unsubscribe at any time.