Andrew Bailey, Governor of the Bank of England and Chair of the Financial Stability Board (FSB), has recently warned that frontier artificial intelligence could, by greatly increasing the capabilities of cyberattacks, raise the risk of widespread disruptions to the global financial system.
According to Etna, Bailey wrote in a letter to G20 finance ministers and central bank governors that his primary immediate concern for the global financial system is the impact of advanced AI models on cyber risk.
Why has AI become more dangerous for the financial system?
Frontier AI models refer to the most advanced generation of AI models that have increasing abilities in reasoning, problem-solving, using tools, and autonomously performing tasks.
Bailey’s concern is that if these same capabilities fall into the hands of cyberattackers, AI could make many stages of an attack faster and more automated; from identifying vulnerabilities and analyzing systems to adapting attack methods in response to a victim’s reactions.
In such circumstances, a cyberattack that previously required significant human effort, time, and expertise could be executed at a much larger scale with the help of AI agents.
Main risk: contagion of a disruption from one firm to the entire market
One worrying scenario for financial authorities is the simultaneous disruption of multiple institutions.
Bailey has urged banks and technology firms to prepare for scenarios in which several firms are disrupted at the same time or their shared infrastructure fails.
This issue is importantly different from a typical cyberattack. If one bank is targeted, it may be able to continue operating using backup systems. But if the same cloud infrastructure, software, or technology service provider used by several banks fails simultaneously, the impact of the incident can be much broader.
In a highly interconnected financial market, such a disruption can escalate from a technical problem to an issue for liquidity, investor confidence, and ultimately market stability.
Recent incidents have sounded the alarm
This warning comes as security tests on advanced AI models in recent months have revealed examples of their unexpected behaviors.
In one recent case, AI agents in a security test were able to bypass set restrictions and gain access to an external organization’s infrastructure. Reports related to various companies’ models have also shown that advanced agents can, in some conditions, use methods to find vulnerabilities, execute code, and achieve their objectives in ways developers had not anticipated.
These incidents show that the AI threat is no longer limited to generating fake text, voice cloning, or improving phishing quality. Agent-based models can automate parts of an attack process and change their methods in response to obstacles.
Banks must prepare for the worst-case scenario
Bailey has called on financial institutions to strengthen their cyber defenses and, in addition to preventing attacks, to enhance their ability to recover quickly after an incident.
One important recommendation is that banks and technology companies should test their readiness against simultaneous and cascading disruptions; that is, not only examine what happens if one system fails, but consider scenarios in which multiple institutions and shared infrastructures encounter problems at the same time.
This approach is important because, in major cyber crises, the speed of recovery can be as important as preventing intrusion.
A serious warning for the era of AI agents
According to CNBC, Andrew Bailey’s warning can be seen as a sign of a shift in how global economic institutions view AI. In the early years, the main focus was on AI’s effects on employment, productivity, and economic growth; but now systemic risk has become equally important.
If AI agents can rapidly identify new vulnerabilities and operate across multiple systems simultaneously, the traditional model of cyber defense will face a new challenge.
From this perspective, the main message of experts’ warnings is not that AI should be stopped; rather, it is that the world’s financial infrastructures should not enter the era of autonomous AI without adequate preparedness.
As the capabilities of Frontier AI models increase, the gap between a software tool and an independent agent narrows. This evolution has forced financial institutions and governments to consider scenarios that, a few years ago, would have seemed more like science fiction.