TL;DR
UK banks have started to pull back from AI-related lending and investment as fears of a sharp correction grow, according to Katharine Braddick, who runs the Prudential Regulation Authority (PRA). Speaking in her first interview since she took over in July, published by Bloomberg on 10 November and reported by FStech on 12 November, she said the Bank of England’s review of bank exposures has not found that lenders would “amplify” an AI shock.
The PRA and the FCA, its conduct counterpart, plan to issue guidance on frontier AI and quantum computing risks later this month, which Braddick described as “quite concrete”.
What the regulator is seeing
“We’re all alarmed at the moment, aren’t we?” Braddick told Bloomberg. The Bank has no current plan to use capital requirements to push banks to cut AI exposure, but she said firms are reviewing it themselves: “I can see firms looking at their risk appetite and thinking and considering, ‘Yes this client, not that client.’ You do see it feed through.”
She called AI risk a “constant conversation” with bank chief executives, alongside operational resilience. The coming guidance will set governance expectations and, she hopes, start an “iterative exercise” with the industry. Bloomberg notes that she joined a few months after Anthropic’s Mythos model alarmed regulators worldwide about AI-enabled cyberattacks.
Her comments sit alongside our recent report on Governor Andrew Bailey’s AI bubble warning.
Resilience and capital
Braddick is also working out how to put Bailey’s “minimum viable bank” idea into practice: core functions that keep running through a cyberattack. Open questions include which banks need infrastructure of their own and how smaller ones could share theirs. “It’s an urgent question,” she said, though no deadline has been set.
On capital, she called a letter from Chancellor John Healey urging faster reforms “completely understandable” but said it had not changed her thinking. The Financial Policy Committee’s decision on capital reforms is due on 26 November. A first five-year PRA strategy is promised early next year.
Looking Forward
For UK lenders, the guidance due this month matters more than the interview’s headline. In our view, it should give a clearer statement of what supervisors expect banks to do about frontier AI as both a credit risk and a cyber risk. Boards with large AI-linked loan books or heavy reliance on AI suppliers should expect questions on both fronts.