The constraint in play is the deployment pipeline itself. Frontier AI models, those operating at the current edge of machine capability, can enter financial infrastructure faster than any supervisory framework has been designed to assess them. Andrew Bailey, head of the world's financial stability watchdog, brought that problem to the G20 directly, warning that AI poses a genuine danger to the global financial system and urging member countries to take concrete steps to control how new frontier models are released.
The mechanism behind that warning is systemic interconnection. Modern financial markets do not operate in silos. A model running at scale in one jurisdiction reaches counterparties across borders, and a failure mode embedded at the frontier of capability can propagate across that network before any human circuit-breaker activates. Financial system speed and cross-border reach are what convert a model-level error into a systemic event.
Financial regulators have tracked AI tools in trading and credit systems for years. Bailey's concern shifts focus to the newest generation of frontier models, where capability has outpaced the frameworks designed to evaluate and contain it.
The release stage as the regulatory target
Bailey's concern, as reported, targets a specific moment in the model lifecycle: the transition from development into live financial systems. That scope is narrower than governing AI broadly, but it still requires coordinated international action to be effective. A frontier model released under a permissive regulatory regime does not stay within that regime's borders. It flows into clearing and settlement systems that cross jurisdictions.
The choice of the G20 as the forum carries its own signal. Financial stability risk is multinational by nature, and routing this concern through the G20 financial track rather than through technology-focused bodies signals that the world's financial stability watchdog now treats frontier AI deployment as part of its core mandate. Bailey's standing in that venue gives the warning institutional weight that a technology forum alone would not.
The definitional challenge remains. "Frontier" is a moving threshold. Any governance framework built around it needs a mechanism to keep pace with capability advances, or it locks in at today's boundary while the actual risk surface moves ahead.