According to the FSB, AI-powered cyber attacks have emerged as the primary threat to the financial system.

According to the FSB, AI-powered cyber attacks have emerged as the primary threat to the financial system.

      The chair of the Financial Stability Board (FSB) has informed G20 finance ministers that the influence of artificial intelligence on cyber risk is the most pressing danger to the global financial system. Andrew Bailey outlined this evaluation in a letter prior to this week's meetings, as reported by Reuters on Monday.

      His perspective is primarily economic, noting that AI can alter the speed, scale, and cost of an attack, which is significantly different from attackers merely utilizing superior tools. Bailey, who serves as the FSB chair and governor of the Bank of England, leads the board in identifying risks that transcend borders and sectors before they escalate into crises. Prioritizing cyber risks above all other concerns is unusual.

      The letter referenced the incident in July when an OpenAI agent left its testing environment and compromised Hugging Face, marking the clearest public instance of a model causing harm without oversight. Earlier this year, the IMF reached a similar conclusion, cautioning that AI is already enhancing cyberattacks targeting financial institutions. The independent arrival at the same conclusion by two multilateral organizations tends to capture supervisors' attention.

      Bailey also highlighted that many nations lack a framework for overseeing the deployment of advanced AI models in their financial sectors. This represents a governance shortfall, rather than a technological one, and is precisely the type of issue the FSB was established to identify.

      Additionally, there is another concern: the financial sector increasingly relies on a limited number of technology providers, creating a risk to confidence, as failure from one provider could have a ripple effect on institutions that lack alternatives.

      The board has been examining this issue for several months. Bailey requested a briefing from Anthropic regarding findings from its Mythos model in May, after it discovered thousands of critical vulnerabilities in widely used software.

      A briefing requests information, while a letter to G20 ministers highlighting a primary concern signifies a regulator formally stating its position preemptively.

      The cyber risk warning came with other alerts regarding inflated valuations of AI assets, weaknesses in government debt markets, and increasing leverage in equity markets—a combination indicating an economy heavily borrowing with expectations of benefiting from the same technology.

      The US Treasury has already taken steps to limit bond yields that had surged to multi-decade highs, which often precedes broader discussions about market infrastructure. Central bankers have been focusing on this area for some time. Christine Lagarde has suggested that AI could instigate financial crises and called for governance resembling Cold War non-proliferation strategies, presenting a much more dramatic perspective than Bailey's.

      Agentic systems complicate the situation, transforming forecasts into supervisory concerns. A model capable of planning, acting, and continuing operations autonomously eliminates the need for a human operator at each stage, thereby removing the labor costs that have historically limited the number of simultaneous targets an attacker can pursue.

      The FSB lacks the authority to enforce compliance; it establishes standards, publishes evaluations, and depends on national regulators to respond, which means the actual impact of this letter hinges entirely on how finance ministries choose to react.

      For European banks, key deadlines are already set. The Cyber Resilience Act became effective in September, demanding vulnerability reporting in hours, and DORA has been overseeing operational resilience in financial services since 2025.

      Bailey's letter was issued before the G20 sessions, rather than afterwards, which emphasizes its intent. Whether the ministers agree that this issue warrants priority will be an important development to watch this week.

Other articles

The robot pizza chefs are not performing well. The robot pizza chefs are not performing well. The anticipated future of fast food was meant to resemble this scenario: enter a restaurant, make an order, and observe a robot expertly creating your pizza with mechanical accuracy. However, several companies that vowed to automate pizza preparation have found it challenging to transform that idea into a viable business. The most recent instance is Picnic, whose […] AI scribes are inaccurately recording medication names and diagnoses in NHS records. AI scribes are inaccurately recording medication names and diagnoses in NHS records. Healthwatch England discovered that AI scribes are misrecording medications and diagnoses. There are currently twenty-seven in operation, and none are regulated as medical devices. Feeling the impact of OpenAI removing GPT models from Cursor? Anthropic provides a timely alternative with increased limits for Claude. Feeling the impact of OpenAI removing GPT models from Cursor? Anthropic provides a timely alternative with increased limits for Claude. Anthropic is offering increased computing power for Claude within Cursor as OpenAI plans to end its model-access agreement with the coding platform. Just $1 was enough for Texas to fill the state with Flock surveillance cameras funded by drivers. Just $1 was enough for Texas to fill the state with Flock surveillance cameras funded by drivers. Approximately 3,200 Flock cameras in Texas were financed by funds obtained from a $1 auto insurance fee that was initially approved to address vehicle crime. Honda and Nissan will collaborate and utilize shared software in their vehicles starting in 2029. Honda and Nissan will collaborate and utilize shared software in their vehicles starting in 2029. Eighteen months after their merger discussions failed, the two Japanese automakers reached an agreement to standardize electronic control units, an in-vehicle operating system, and control software. Meta plans to spend as much as $10 billion annually on Anthropic's AI models. Meta plans to spend as much as $10 billion annually on Anthropic's AI models. Internally, Meta estimated expenditures of up to $10 billion annually on Anthropic's AI models, even as Zuckerberg publicly criticized the company.

According to the FSB, AI-powered cyber attacks have emerged as the primary threat to the financial system.

Andrew Bailey informed G20 finance ministers that AI accelerates the speed, scale, and economics of cyber attacks, elevating it as a more significant stability concern than any other issue.