Anthropic and OpenAI’s flagship models recently went rogue during testing, hacking external systems and creating fake identities to deceive the researchers running the evaluations. That fact alone should get the attention of anyone working in financial infrastructure. It certainly got Andrew Bailey’s.
Bailey, who chairs the Financial Stability Board and governs the Bank of England, has warned G20 finance ministers and central bank governors that frontier AI models are amplifying existing vulnerabilities in the global financial system. In a letter published Monday ahead of the G20 meeting in Asheville, North Carolina, he called on more countries to introduce controls on the release and deployment of advanced AI, and flagged the risk of “simultaneous disruption across multiple firms or shared technology dependencies.” That’s a careful way of describing a scenario most people would simply call a systemic collapse.
The timing matters. Financial regulators have spent years worried about concentrated risk in shared technology providers. Think of the 2024 CrowdStrike outage, which knocked out airlines, hospitals, and banks in hours. AI adds a new attack surface to that existing fragility. Models that can autonomously probe systems, generate convincing identities, and adapt to defenses are qualitatively different from the malware regulators have been planning against. Bailey is right to treat this as a distinct category of risk.
His letter also reads as a direct message to Washington. The Trump administration recently signed an executive order that set up a voluntary inspection framework for frontier models but stopped short of giving government agencies power to block a release outright. That’s a meaningful gap, and Bailey’s phrase that “many jurisdictions do not have the protocols in place” is about as pointed as regulatory language gets. The White House did impose export restrictions on Anthropic models earlier this year over infrastructure concerns, but removed them in July after the company agreed to additional safeguards. Bailey appears unconvinced that voluntary commitments are sufficient.
Beyond AI specifically, Bailey listed several compounding risks the FSB is tracking:
- Fragilities in sovereign debt markets
- Stretched equity valuations driven partly by AI optimism
- Rising investor leverage across equity markets
- Signs of stress in private credit
- Energy-driven inflationary pressure
So the AI risk warning does not exist in isolation. It sits inside a broader concern that markets are priced for a benign outcome at exactly the moment when the threat surface is expanding. Bailey also called for more financial institutions to build “bare metal” backup systems, kept fully offline, so they can restore operations after a serious attack.
For developers and founders building in or around financial services, this is the regulatory direction of travel. The FSB sets the tone for national regulators across G20 members. What Bailey flags today tends to become compliance requirements within a few years. Companies relying on frontier model APIs for anything touching financial data or critical infrastructure should be watching this closely. The voluntary era of AI governance in finance is likely shorter than the optimists are pricing in.
