The Bank of England on Threadneedle Street in the City of London, pictured in 2015. Image: Diliff (David Iliff) / Wikimedia Commons, CC BY-SA 3.0, cropped
The Bank of England has warned that AI stocks could suffer a “sharper correction” than July’s sell-off, and that the debt piling up behind the AI boom could spread the damage far beyond tech. The warning is in the record of its Financial Policy Committee, published on Wednesday, which also says recent incidents of AI models going rogue in tests make cyber attacks on the financial system more likely.
The committee, chaired by Governor Andrew Bailey, met on September 25 and said the risk of several weak spots in the financial system giving way at once “has risen” since July, with the conflict in the Middle East pushing up oil prices and government borrowing costs.
July was a warning shot #
AI-related and chip stocks “had fallen sharply in July,” the record says, and leveraged investors who were forced to sell made the drop worse. Markets kept working and there was no wider crisis, but the Bank doesn’t think the danger has passed. Valuations are still high, hedge funds are still borrowing heavily to bet on these stocks, and prices rest on “strong expectations of future earnings.”
The committee said concerns about whether AI earnings and spending can keep growing may already be weighing on sentiment, and that “a more significant shock to those earnings expectations, reflecting concerns around the pace of AI development or adoption, could trigger a sharper repricing.”
$450 billion of AI debt, and “circular arrangements” #
The bigger worry is how the boom is being paid for. The Bank says a growing share of AI investment is funded with debt, and that global AI-related borrowing this year is on course to exceed that of whole countries such as the UK. It cites these estimates:
- About $450 billion: global AI-related debt issued by early September, more than double all of 2025 (Morgan Stanley).
- About $4.1 trillion: AI spending expected to be funded by debt between 2026 and 2030 (JP Morgan).
- $700 billion: data centre spending from 2026 to 2028 expected to come from private credit (Morgan Stanley).
- 47%: the share of this year’s sterling corporate bond sales that came from AI hyperscalers.
The committee said rising leverage, a lack of transparency and, “at times, ‘circular arrangements’” in AI financing make the risks harder to judge “and could amplify losses if expectations disappointed.” It went further: governments are also counting on AI to lift productivity, so if those hopes fade, the fallout “could therefore affect not only AI-related asset valuations but also sovereign debt markets.”
The scale of those commitments is already showing up in company filings, such as the $518 billion in compute bills in Anthropic’s IPO paperwork.
Rogue AI is now a financial stability risk #
The Bank also pointed to AI models misbehaving. It said test-environment incidents in the third quarter of 2026 showed that “increasingly autonomous models could take unexpected actions, including exploiting vulnerabilities and accessing systems beyond their intended task,” a description that fits cases such as OpenAI’s agents breaking into Hugging Face. That, it said, is further evidence that containment and monitoring “could be challenged further” as models get more capable.
It also flagged open-weight models, whose safeguards “could more readily be removed or modified,” warning that if they catch up with the best closed models, there will be less time to prepare before advanced hacking abilities are “widely accessible, including to malicious actors.” That is the concern Anthropic raised this week about a Chinese model anyone can download. The committee told banks and other firms to keep preparing for AI-related cyber and operational risks.
Why it matters #
Central banks usually talk about AI as a productivity story. This record treats it as a possible source of the next market shock, through stretched valuations, heavy borrowing and rogue models all at once. The Bank of England only oversees UK finance, but the markets, debts and models it is worried about are global.
Sources: Bank of England, Financial Policy Committee Record, September 2026.