.svg "Group 2238 (2)"){kind=small}
.svg)
.svg)
%20(1).svg)
%20(1).svg)
.svg)
Share Article
Chain three AI agents together, each individually 80% accurate, and the combined system gets it right just 52% of the time — exactly the kind of math regulators from the EU, UK, Singapore, Japan, Germany, and the BIS spent this panel wrestling with. Their real question: when an autonomous agent makes a costly mistake, who actually answers for it?
The EU's AI compliance delay is a warning, not a reprieve
Brando Benifei said the EU's "AI Omnibus," approved by Parliament just the week before the panel, pushed back high-risk AI obligations for financial services (including credit scoring and insurance pricing) from this August to December 2027 for standalone systems and August 2028 for AI embedded in regulated products, giving institutions more runway to build proper governance. But he was blunt that the extra time was meant as "an implementation window for meaningful compliance," not an excuse to wait, and that responsibility for AI literacy stayed primarily with the institutions that built and deployed these systems, not the Commission, since human oversight is meaningless if the people supervising an AI system don't understand its data, its limits, and when to challenge its output.
Who is actually on the hook when an AI agent makes a costly mistake?
Ian Phoenix said the UK's Financial Conduct Authority wasn't writing a new rulebook for agentic AI at all, relying instead on its existing Senior Managers Regime: however much autonomy a firm hands to an AI agent, accountability still sits with the board and senior management, not the model developer or the regulator. He flagged a sobering stat from Gartner: chain three AI agents together, each individually 80% accurate, and the combined chain's accuracy drops to just 52%. Kenneth Gay said the Monetary Authority of Singapore takes the same accountability stance, but pointed to a shift already underway: sophisticated financial institutions are telling MAS their own clients are now asking "where are your agents, so they can talk to my agents," meaning demand for agentic finance is coming from customers, not just built speculatively by firms.
Speakers:
Brando Benifei, Member, European Parliament
Ian Phoenix, Director of Digital & Intelligence, Financial Conduct Authority
Kenneth Gay, Chief FinTech Officer, Monetary Authority of Singapore
Prof. Dr. Fritzi Köhler-Geib, Member of the Executive Board, Deutsche Bundesbank
Ryosuke Ushida, Director for International Policy, International Affairs Office, Financial Services Agency of Japan (JFSA)
Tommaso Mancini-Griffoli, Head of the Innovation Hub, BIS Innovation Hub
Host:
Joasia E. Popowicz, Associate Editor for the Central Banking Publications
-1.jpg?width=352&name=Artboard%201%20(2)-1.jpg)