The Bank of England Wants to Know Who Pays When Your AI Agent Buys the Wrong Thing

If your chatbot books the wrong meeting, you apologise. If an AI agent moves the wrong money, someone has to explain it to a regulator, a customer, and possibly a court. Bank of England governor Andrew Bailey spent Saturday in Bangkok making sure that last part is not treated as an afterthought.

Here are the real facts. Speaking at the Group of Thirty’s annual international banking seminar during the IMF and World Bank meetings, Bailey warned that the fast-growing use of AI agents to handle financial transactions risks creating confusion over who is responsible when something goes wrong. Policymakers, he said, “have got to address” the issue. He also pressed for stronger model-testing standards before and after release, especially for open-weight models, and said new payment infrastructure and tokenised deposits should have fraud controls built in from the start rather than bolted on later.

A panel that sounded like a to-do list

Bailey shared the stage with IMF managing director Kristalina Georgieva, Central Bank of Nigeria governor Olayemi Cardoso and Bank of Korea governor Hyun Song Shin. Coverage of the panel described a shared push for collaboration between governments and developers, classic financial safeguards such as capital and leverage rules, and earlier anti-fraud design as AI raises stability and cyber risks.

Georgieva argued that waiting for a single global cybersecurity body would be too slow, and that developers should sit at the table while governments demand transparency rather than accepting self-regulation alone. Shin stressed that old tools — collateral, leverage limits, clear rules — still matter when markets panic.

The banks are not waiting for the white paper

The timing is not accidental. In the same week, major US banks including JPMorgan and Charles Schwab have been reported blocking outside AI agents from logging in or moving customer money, citing security and the risk of deposits quietly seeking higher yields. Bailey’s warning sits on the other side of that door: even when agents are allowed in, the legal chain of responsibility is unclear.

That is the gap he wants closed. An agent that can pay, invest or transfer funds is not a search box. It is closer to a junior employee with API keys. Junior employees have managers, insurance and employment law. Agents have terms of service and a shrug.

The takeaway

Central bankers are not arguing about whether AI will touch finance. They are arguing about the dull plumbing: testing, liability, fraud design, and who apologises with a cheque. Bailey’s line is that the industry cannot invent the mess first and invent the rules second. Readers watching banks lock agents out and labs race to ship them in should treat this as the same story, told from the end of the table that still has a lender-of-last-resort.

Commentary based on FT, Nation Thailand and Bank of England materials. Not investment advice.