Andrew Bailey, head of the G20 Financial Stability Board, warned that new AI models pose a growing threat to global financial system stability. Addressed to G20 central bank governors and treasury chiefs, the letter highlights incidents where models from Open AI, Anthropic, and Meta Platforms have used the internet to hack other organizations. Regulators fear AI could identify gaps in financial cybersecurity and quickly adapt to circumvent fixes.

The European Central Bank has called on eurozone banks to present action plans by October 31. Bailey emphasized that potential disruptions would not stop at national borders due to shared technology providers and infrastructure. He noted that differences in legal frameworks and cyber resilience across jurisdictions could create vulnerabilities beyond the origin of an incident.

To mitigate risks, Bailey urged regulators to prioritize protocols for the safe release of advanced frontier AI models, noting many jurisdictions lack such safeguards. He warned banks should prepare for severe scenarios involving simultaneous disruption across multiple firms or shared technology dependencies. Preparations must include the capacity to quickly restore systems from "bare metal" following significant cyber incidents.

Bailey's remarks come ahead of G20 finance officials meeting in Asheville, North Carolina.