The central bank warns cutting edge AI could power sharper cyber attacks on banks and trigger an AI stock correction that hits UK GDP by 2.2 percentage points, while easing post 2008 leverage rules.
Britain's financial regulators have made a deliberate rotation. In the same week, the Bank of England's Financial Stability Report classified AI-related financial stability risks as 'significant' for the first time, and its Financial Policy Committee opened a consultation on letting the country's largest banks hold less capital against their loans. To a non-beat reader, the two moves look like mixed signals: alarm on AI plus easier rules for banks. They are consistent within the FPC's own framing.
The FPC is rotating attention from where its existing capital buffers already work to where they cannot. The leverage ratio limits how much a bank can borrow against its capital, and was built after 2008 to absorb loan losses on a balance sheet. The risks the FSR names this round are AI-driven cyber attacks on bank and market-plumbing software, and an AI-stock correction transmitted through equity markets into UK GDP. Those channels are not the ones a leverage buffer catches. Easing leverage while elevating AI risk is consistent because the FPC has, in its own words, judged bank-balance-sheet risk to have receded and AI risk to have grown.
The new pressure point is frontier AI, the most capable general-purpose AI systems. The FSR says these models are increasingly able to exploit software vulnerabilities and 'could increase the sophistication and impact of cyber attacks on firms,' including banks and the market plumbing they rely on. The mechanism is concrete. A frontier model can probe a bank's perimeter software for unpatched weaknesses or generate phishing aimed at employees with network access. The FPC has not said a frontier AI has done any of this to a UK lender; it has said the capability now exists and bank security teams need to keep up. The report also sizes the market channel. In the FPC's first published AI-stock stress scenario, a sharp correction in US-listed AI equities could spill into UK markets and reduce UK GDP by up to 2.2 percentage points. The figure is a hypothetical, not a forecast; the FPC built it to measure the spillover.
The capital easing targets a different risk. The FPC's own stress work judges UK banks as having rebuilt their post-2008 buffers, and treats the leverage ratio as tighter than the underlying risk now warrants. The committee proposes to cut the leverage requirement for large UK-domestic lenders by around 20 basis points (a basis point is one hundredth of a percentage point), with the exact cut varying by bank. That step is the FPC's proposed outcome from a review it opened in response to UK bank arguments that the leverage rule was binding without adding safety. The AI and cyber risks named in the FSR sit outside the leverage buffer's protective scope; the FPC's stress work has not flagged them as balance-sheet risks.
Private credit is one of the 'more pronounced' vulnerabilities the FSR carries forward. The Bank flags continued growth in lending by non-bank vehicles to companies that would previously have borrowed from a bank, with less disclosure than bank balance sheets carry. A sharp AI correction, or a cyber attack that disrupts market infrastructure, could trigger redemptions and refinancing stress in funds that hold those loans. Private credit is the channel that connects an AI-stock shock to UK lenders that did not own the AI stocks.
The same FSR says risks to UK financial stability have increased in 2026 but stresses that UK lenders and consumers remain 'resilient.' Vulnerabilities flagged since the December 2025 FSR have grown more pronounced. AI-linked share prices have risen further; the FSR notes that 'on some metrics, valuations have also become more stretched.' Sovereign debt vulnerabilities, inherited from prior reports, are also flagged as more pronounced in this round. The Bank cites the Middle East conflict as a separate, compounding source of macro uncertainty that could amplify market stress.
The July 2026 elevation is the crossing of a threshold the FPC set itself. Its April 2025 feature on AI in the financial system framed AI as a financial-stability concern to test, and the April 2026 FPC record had already begun treating AI risk in operational and cyber terms rather than only market-valuation language. Calling it 'significant' this round marks the language change the FPC's prior work was building toward, not a sudden reaction to a news cycle.
The leverage-ratio consultation closes later this year. The FPC's next AI stress round feeds into the December 2026 FSR, the natural test of whether the new 'significant' classification produces new supervisory expectations or stays a status-quo flag. Wire coverage of the July report, in the Independent and Leader Live, has tended to treat the AI warning and the capital easing as separate items. The Bank's own publication runs both in the same document.