The Payment Expert Podcast examined the Financial Stability Board’s final agentic AI guidance, the FCA’s plan to become an “agentic regulator”, and what AI agents acting for consumers mean for fraud and anti-money laundering controls.
The Financial Stability Board (FSB) published 12 practices for financial institutions using agentic AI on 15 October, covering governance, data and human oversight.
The practices are not binding but will go to the G20, the podcast discussed, and a standardised code is needed as agentic commerce and payments grow.
The FSB’s final report on responsible AI adoption states that AI agents do not reason like humans and may interpret predefined goals in unexpected ways.
The team discussed why agents should not be treated as “synthetic employees”. That framing implies an AI agents can be blamed when something goes wrong, when responsibility should sit with the person overseeing it.
The practices apply to financial institutions, not to the tech and payment firms building their own agentic AI standards. PSPs may still favour providers that have built around the practices, the podcast said.
FCA plans AI-led supervision
Jessica Rusu, the Financial Conduct Authority‘s (FCA) Chief Data, Information and Intelligence Officer, set out the regulator’s AI plans in a speech on 19 October. She said the FCA is developing “our future as an agentic regulator.” The FCA is building AI tools to supervise around 60,000 anti-money laundering firms entering its remit in 2028.
The FCA is not writing AI-specific rules. The podcast said that approach holds because rules would be obsolete by the time they were drafted and implemented. Consumer Duty covers authorised firms but not general-purpose AI agents, which leaves a gap, the team said.
The FCA’s poor practice examples included using AI without a defined purpose, excluding vulnerable consumers by default and overstating capabilities. Its good practice examples included clear consumer choice and access to a human.
Fraud detection without human patterns
Human oversight must be enforced, the podcast said, with people able to stop a rogue agent from completing a transaction.
Fraud and AML controls will need to be rebuilt, the team discussed. Banks and payment firms rely on transaction monitoring that compares payments with a customer’s habits to detect fraudulent card purchases.
AI agents can transact at any hour and show no such pattern, the podcast said, which makes an unauthorised or fraudulent agent purchase harder to spot.
The latest episode of the Payment Expert Podcast is available now on Spotify, YouTube, Apple Podcasts, and all major podcast platforms. Subscribe to the Payment Expert newsletter to make sure you don’t miss an episode.
