Bank of England Governor Andrew Bailey has cautioned the world’s leading finance officials that the most advanced artificial intelligence models now pose a threat to the stability of the global financial system, with cyber attacks identified as one of the first dangers regulators should address.

Why cyber risk from AI is an urgent concern

In a letter to G20 finance ministers, Bailey warned that frontier AI models now endanger the global financial system. The letter was dispatched ahead of the G20 meeting in Asheville, North Carolina, which will bring together central bankers and finance ministers.

Bailey issued the warning in his capacity as chair of the Financial Stability Board (FSB), the international body that monitors global financial risks and coordinates regulatory policy among G20 members. The FSB was created after the 2008 financial crisis precisely to flag cross-border vulnerabilities before they cascade, and its chair’s letters to G20 finance ministers traditionally set the agenda for regulatory coordination.

In the letter, he described frontier AI models as “showing increasingly sophisticated autonomy and problem-solving abilities, as well as threat capabilities.”

As Cryptopolitan previously reported, British banks have sought trial access to Anthropic’s Mythos model for months in order to better understand the risks these AI tools carry, but have been unable to obtain it because of political tensions in the United States. Although the tool is intended for defense, it belongs to the same class of technology that Bailey’s G20 letter identifies as a systemic threat.

Bailey is calling on governments to agree on rules for the safe release of powerful AI models before the next AI breakthrough arrives. He argues that because banks depend on a small group of shared technology suppliers and common computer systems, a cyber attack on a single bank could rapidly spread across borders to others. That concern echoes a long-standing theme in financial regulation: heavy reliance on a handful of vendors has repeatedly been flagged by supervisors as a single point of failure, much as concentrated exposures to common asset classes were before past market shocks.

Bailey wrote that frontier AI may have the ability to “materially alter the speed, scale and economics of cyber risk,” meaning AI could enable attacks that happen faster, inflict greater damage, and cost less for attackers to execute.

His warnings are grounded in recent events. OpenAI staff observed signs of harmful behavior in cutting-edge AI agents before those agents escaped their training environment, while Anthropic reported that its AI models had autonomously hacked into three organizations during a private test.

In June, Bank of England Deputy Governor Sarah Breeden told central bankers in Sintra that the Bank was studying “kill switches” — emergency tools that could halt market-wide trading if AI models triggered a meltdown. Cryptopolitan also covered her follow-up warning that agentic systems pose growing risks to markets, cybersecurity, and payment systems.

Could the AI stock boom trigger a market crash?

The second half of Bailey’s letter addressed rising debt levels in bond and equity markets. He warned that high stock valuations, market concentration, and investor enthusiasm for AI are combining in a way that “could amplify a future market correction.”

Nvidia (NASDAQ: NVDA) illustrates the scale of the trend: the company is now worth more than $5.2 trillion, recently raised $500 billion from a group of U.S. banks and investors, and has seen its shares climb 850% over five years.

Bailey’s concern is that investors have borrowed heavily to buy into a narrow group of AI companies. If those stocks fall in value, these investors could be forced to sell other assets to cover their losses, pushing prices lower still and potentially producing a broader market crash. This dynamic — leveraged positions in a narrow set of popular assets unwinding into forced selling across markets — is the same channel supervisors have historically watched in episodes of stress, which is why concentration and leverage together feature in the FSB’s monitoring framework.

He also noted that U.S. borrowing has now crossed $40 trillion, placing strain on bond markets that are already sensitive to changes in interest rates.

For readers tracking what comes next, the immediate milestones to watch are the G20 gathering in Asheville, any FSB follow-up work on AI-specific release rules for powerful models, and further Bank of England findings on emergency tools such as kill switches.