He said the US government and AI companies are coordinating after he and Federal Reserve Chair Jerome Powell held a flash meeting with Wall Street bank chiefs.
The talks focused on cyber risks linked to Anthropic's Claude Mythos Preview, which Bessent said reflected a recent step change in large language model capabilities.
He said Washington must balance safety with innovation and keep ahead of China, as regulators, banks and AI firms strengthen financial-system defences against faster, larger-scale attacks.
As AI automates cyberattacks, who is liable when a banking AI autonomously causes a breach?
With AI now able to find decades-old security flaws, is any financial data truly safe?