Bank of England’s Bailey warns frontier AI raises cyberattack risks

Andrew Bailey has a new item on his worry list, and it is not inflation. The Governor of the Bank of England says frontier AI models increase the risk of cyberattacks. He wants those systems tested more rigorously, both before and after they are released.

Bailey also chairs the Financial Stability Board (FSB), so this is not one central banker musing aloud. When the person in charge of coordinating global financial stability sounds an alarm, regulators in other countries tend to take notes.

What Bailey is actually saying

The core argument is about speed, scale, and cost. Bailey claims frontier AI significantly increases how fast cyberattacks can move, how many targets they can reach, and how cheap they are to run.

A joint statement from the Bank of England, the Financial Conduct Authority (FCA), and HM Treasury put the concern in plain terms. These AI systems can now handle tasks that once required skilled human operators.

That matters because skill used to be a bottleneck. Sophisticated attacks needed sophisticated people, and those people are scarce. If software can do the work, the pool of potential attackers widens considerably.

According to the research findings, as of May 15, 2026, frontier AI models have shown cyberattack capabilities that surpass those of skilled human professionals. The Bank of England has pointed to an April 2026 evaluation of OpenAI’s GPT-5.5 cyber capabilities as part of the evidence behind its warnings.

Bailey has framed the shift in a letter to G20 ministers, writing that frontier AI:

“may materially alter the speed, scale, and economics of cyber risk”

Why the financial system is especially exposed

He has warned that malicious uses of these models threaten the safety of single firms and the integrity of the broader market. The potential systemic effects, he argues, are made worse by shared dependencies among financial institutions.

The Bank has also been working with the AI Security Institute and the National Cyber Security Centre (NCSC). One focus is the higher volume of vulnerabilities that now need faster patching.

The Bank is encouraging firms to strengthen their cyber resilience, with emphasis on three capabilities: detection, containment, and recovery.

The policy asks

Bailey’s proposed remedies go beyond telling banks to buy better firewalls. He is advocating for rigorous testing of AI models before deployment and continued testing after release.

He has also called for international collaboration on safety protocols. That fits his FSB role, since cyber threats do not stop at borders and neither do the AI models enabling them.

Perhaps most notably, Bailey has backed establishing a societal right to intervene in the deployment of such technologies. That is a significant statement from a central bank governor. It suggests the public, through regulators or governments, should retain some ability to step in rather than leaving release decisions entirely to developers.

In an opinion piece, Bailey called for regulatory frameworks that evolve alongside AI, describing the risks as real and increasingly significant.

The broader backdrop

Bailey has repeatedly pressed for frameworks that can adapt as the technology develops. His argument is that the financial system needs to be ready for a rising volume of vulnerabilities created by AI advances.

Bailey has also tied AI to financial stability concerns beyond cybersecurity. In statements during 2026, he connected AI-related risks to potential instability, noting that AI-driven debt growth could surpass $450 billion by early September 2026.

What this means for banks, investors, and AI developers

For financial institutions, the message is fairly direct: cyber resilience is moving up the supervisory agenda. When a central bank, a conduct regulator, and the Treasury issue a joint statement on the same risk, budget committees tend to notice.

For AI developers, the testing push is the headline risk. Pre-release evaluations are already becoming common practice for frontier labs. Bailey’s call for ongoing post-release testing, plus a societal right to intervene, would represent a meaningful step further.

The thing to watch next is whether other central banks and financial regulators echo the language. If the FSB starts building AI cyber risk into its formal work, the conversation shifts from one governor’s warning to a coordinated supervisory priority.