Bank of England Governor Andrew Bailey is warning that cyber risks from frontier AI extend beyond faster attacks: the technology can also make them cheaper, expand their reach and put sophisticated capabilities in more hands. He is calling for tougher testing of AI models before deployment and after release, alongside a societal right to intervene in deployment decisions.
Key takeaways
- Bailey says frontier AI can lower the barriers to sophisticated cyberattacks.
- The Bank of England cites an April 2026 evaluation of OpenAI’s GPT-5.5.
- His proposals include continued model testing and international safety cooperation.
- Shared dependencies among financial firms can amplify systemic threats.
How frontier AI increases cyber risks
Frontier AI can accelerate cyberattacks, increase the number of targets they reach and reduce their cost, Bailey says. A joint statement by the Bank of England, Financial Conduct Authority and HM Treasury highlights another change: models can perform work that previously demanded skilled human operators.
That capability widens the potential attacker pool. Tasks once constrained by the availability of technically skilled people can increasingly be handled by software.
Bailey, who also chairs the Financial Stability Board, described the implications in a letter to G20 ministers. Frontier AI, he wrote, “may materially alter the speed, scale, and economics of cyber risk”.
The evidence behind the warning
As supporting evidence for its concerns, the Bank of England pointed to an April 2026 evaluation of OpenAI’s GPT-5.5 cyber capabilities. The research findings described in the warning indicated that, as of May 15, 2026, frontier models had demonstrated cyberattack capabilities exceeding those of skilled human professionals.
The concern therefore encompasses both the performance of the models and access to that performance. Bailey’s warning links stronger technical capabilities with the ability to conduct attacks more quickly and cheaply.
Testing before deployment—and after release
Bailey’s proposed response to cyber risks from frontier AI includes rigorous testing before and after release, rather than treating deployment as the end of the evaluation process. He has also called for countries to cooperate on AI safety protocols.
His proposals extend to establishing a societal right to intervene in the deployment of these technologies. In an opinion piece, he advocated regulatory frameworks that keep pace with AI’s development, describing the risks as real and increasingly significant.
The Bank is also working with the AI Security Institute and National Cyber Security Centre. Their work includes addressing the growing volume of vulnerabilities requiring faster patching.
Financial stability concerns extend beyond individual firms
Bailey warns that malicious uses of frontier models threaten individual financial firms as well as the integrity of the wider market. Shared dependencies among institutions, he argues, can magnify the systemic effects.
His concerns also reach beyond cybersecurity. In statements during 2026, Bailey linked AI-related developments to potential financial instability, citing a projection that AI-driven debt growth could exceed $450 billion by early September 2026. That figure was a projected risk, not a confirmed total.
For financial institutions, the Bank’s stated resilience priorities are detection, containment and recovery, alongside preparation for the rising volume of vulnerabilities associated with AI advances.
Read also: Avalanche could underpin NYSE-linked tokenized securities platform, pending approval · Dinari plans tokenized Anthropic shares for IPO day, with U.S. access
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.