Bank of England Demands Right to Intervene as AI Debt Hits $450B and Cyber Risk Climbs

Britain's top central banker wants a hand on the brake. Bank of England Governor Andrew Bailey said on Wednesday that society must retain the right to intervene in frontier AI, warning that the risks from the most powerful models are real and increasingly significant.

He made the case as the Bank flagged a second worry closer to its core job: the AI boom is being financed with a rapidly growing pile of debt, and that debt is now tangled up with the rest of the financial system.

$450 billion in nine months

The Bank's Financial Policy Committee, which met on September 25, said AI-related borrowing is raising financial stability risks. Large players took on $450 billion between January and September 2026.

To put that in perspective, it already exceeds the $333 billion of UK government bonds due to be issued for the whole of 2026. The money ties hedge funds, asset managers and private credit funds to AI companies that have yet to turn a profit. If the expected returns do not show up, the losses will not stay inside Silicon Valley. That concern echoes the debate over whether Big Tech's AI capex is running ahead of earnings.

Cyber threats that could hit payments

Bailey's other warning is about attacks. In an opinion piece for the Bank's Insight series, he said AI has increased the scale and sophistication of cyber threats capable of disrupting everyday card payments, bank transactions and trading in stocks and bonds.

He also flagged a quieter oversight problem. When powerful models operate in a self-reinforcing loop, he argued, it becomes harder for the public to see what they are doing and to step in. That is the core of his demand: authorities need to keep the ability to set boundaries and revise them as the technology changes.

Test first, regulate later

Bailey was careful not to call for an immediate rulebook. He said the potential benefits of AI are immense, and that jumping straight into arguments about regulatory architecture is not the right place to start.

His suggested first step is rigorous testing to find credible points where humans can intervene in how a model behaves. Those intervention points could later harden into standards. It is an engineer's approach to a political problem: work out where the controls could go before deciding who holds them.

The comments build on a letter Bailey sent to G20 leaders in August as chair of the Financial Stability Board, which put frontier AI and cyber risk on the global agenda. Wednesday's piece speaks for the Bank of England and lands alongside hard numbers from its own committee.

Why markets should listen

Central bankers rarely weigh in on technology governance. Bailey doing so, while his colleagues put a dollar figure on AI leverage, signals that the Bank now sees the AI trade as a potential source of systemic stress rather than just a growth story.

The Bank has not moved to restrict lending to AI firms. But it has said publicly, in plain terms, that it is watching both the debt and the models, and that it wants the power to act on either.