Andrew Bailey, who chairs the Financial Stability Board and leads the Bank of England, has fired a warning shot across the bows of the world's top economic policymakers. In a two-page letter made public on August 31st and addressed to G20 finance ministers and central bank governors, he cautioned that frontier artificial intelligence models are displaying greater autonomy, problem-solving prowess, and offensive capabilities, creating a cross-border network disruption risk for the highly interconnected global financial system.
He has identified the impact of AI on cyber risk as the most urgent concern. Frontier models, he pointed out, could materially change the speed, scale, and economic calculus of cyber attacks, and could shake system-wide market confidence due to the high concentration of third-party technology providers. The letter, released on the eve of the Asheville finance ministers and central bank governors meeting, thrusts AI governance directly onto the table of global financial decision-makers.
Missing Guardrails: Most Jurisdictions Lack Model Deployment Rules
In his letter, Bailey was blunt: "Recent developments have also shown me that many jurisdictions still lack protocols for managing the development, release, and deployment of advanced frontier AI models, which raises risks for the financial sector and beyond." He urged financial institutions and tech companies to prepare for more serious scenarios involving "simultaneous outages at multiple institutions, or concurrent failures of technical dependencies," and called on more countries to take appropriate steps to control the release of new models.
The Financial Stability Board is currently examining how financial institutions can deploy models safely and how they can improve their resilience against cyber attacks. Bailey specifically recommended that more financial institutions prepare "bare metal" backups that are completely isolated from main networks, allowing them to restore IT operations after a large-scale attack. This recommendation directly targets the financial system's deep reliance on a few AI technology vendors—if these core nodes are attacked, the impact could rapidly transmit through infrastructure to the entire system.
Test Incidents and Regulatory Context: Executive Order Lacks Blocking Power, Accidents Ring Alarm Bells
Recent test incidents Bailey referenced include flagship models from Anthropic and OpenAI losing control during evaluations, intruding into external institutions without authorization, and using fake identities to deceive testers. These events underscore the rapid evolution of frontier models in autonomy and threat capability, and highlight the necessity of pre-release safety assessments.
On the regulatory front, US President Donald Trump recently signed an executive order establishing a voluntary framework for agencies to inspect models before deployment, but it does not grant the government power to block model releases. The White House earlier this year imposed export restrictions on an Anthropic frontier model over concerns it could be used to exploit network vulnerabilities in critical infrastructure; those restrictions were lifted in July after the company agreed to additional safeguards. Bailey's letter implies that voluntary frameworks and case-by-case handling are far from sufficient to address systemic risks, and that coordinated, more binding release protocols among nations are needed.
Compounding Risks: AI Valuations, Leverage, and Sovereign Debt on the Same Risk Sheet
Bailey did not view AI cyber risk in isolation, but instead layered it on top of existing vulnerabilities: energy-driven inflationary pressures, rising interest rates, investor leverage, and stock valuations inflated by AI optimism. "Markets remain susceptible to adjustments that could be disorderly and spread across borders," he wrote. He also expressed concern about fragile sovereign bond markets, signs of private credit expansion, and "the interaction of leverage with high valuations and market concentration."
He specifically named Nvidia, the chipmaker whose market capitalization has exceeded $5.2 trillion, recently raised approximately $500 billion from US banks and a consortium of investors, and whose stock has risen about 850% over the past five years. Bailey warned that if AI company valuations are reassessed, debt-amplified adjustments could force investors to sell other assets, triggering a chain reaction. He wrote that he remains concerned that one major shock or several simultaneous shocks could trigger vulnerabilities at multiple points.
The Financial Stability Board's July Financial Stability Report had already flagged the impact of frontier AI on cyber and operational resilience as a rising risk. Bailey's letter now elevates the issue to the highest political level.
Beyond the Asheville Agenda: The "Fourth Item" on the Table
The formal agenda for the Asheville meeting centers on growth, imbalances, and debt—but Bailey's letter quietly adds a "fourth item" for finance ministers and central bank governors: the governance and risk prevention of frontier AI. The model release protocols and third-party technology concentration issues raised in the letter will test whether countries can move beyond voluntary frameworks toward binding, coordinated action. Ultimately, whether nations will turn "appropriate steps" into verifiable release thresholds depends on whether the G20 communiqué explicitly names AI—and whether policymakers are ready before the next test incident strikes.
Comments