New artificial intelligence models pose an increasing threat to the stability of the global financial system.
Ensuring their safe deployment must be a priority, the head of the Financial Stability Board (FSB) stated in a letter to G20 regulators published on Monday.
Andrew Bailey, also Governor of the Bank of England, sent the letter to central bank governors and finance ministers of G20 member states. The letter follows a series of recent incidents in which new models from companies including OpenAI, Anthropic, and Meta Platforms (NASDAQ:META), which is designated an extremist organization and banned in Russia, were used to hack other organizations online.
Regulators are concerned that new AI models are capable of identifying previously unknown vulnerabilities in financial institutions' cybersecurity systems and quickly adapting, bypassing newly implemented defenses. The European Central Bank has ordered eurozone banks to submit an action plan to address the heightened risks associated with new AI models by October 31, 2025.
"The risk environment has been further complicated by the emergence of advanced AI models, which demonstrate increasingly sophisticated autonomy and problem-solving capabilities, as well as the potential to create threats," Bailey wrote.
A meeting of G20 finance ministers is scheduled for that same Monday in Asheville, North Carolina. The FSB serves as the coordinating body for financial regulators in the G20 countries.
Bailey warned that the impact of AI-induced disruptions will not be limited to national borders, given the shared technology providers and infrastructure that underpin the global financial system. "Differences in legal frameworks, cybersecurity capabilities, resilience, and resilience across jurisdictions can have implications far beyond the jurisdiction where an incident occurs and can themselves be a source of vulnerability," he wrote.
To limit the risk of an AI-enabled cyberattack spreading through the global financial system, Bailey urged regulators to prioritize developing protocols for the secure release of new models. "Many jurisdictions lack protocols for managing the development, release, and deployment of advanced AI models, increasing risks for the financial sector and beyond," he noted.
Bailey also added that banks and other financial institutions should prepare for "more severe scenarios involving simultaneous disruptions to multiple companies or shared technology dependencies." Such preparation, he said, should include the ability to quickly restore computer systems after significant damage. "These events highlight the importance of robust response and recovery capabilities, including the ability to restore critical systems and data from scratch after a significant cyber incident," Bailey said.