The White House is reportedly considering vetting frontier AI models before release in response to growing security risks.
The Trump administration has been non-interventionist on AI, but Anthropic’s new system Mythos proved capable of finding security flaws in almost all websites, and the firm has held back its public release as a result: The EU said it was in contact with Anthropic to get the region’s banks tested for security vulnerabilities.
Other models will soon be similarly capable, and their developers may not be so cautious. The White House plan, modeled after Britain’s approach, would have officials review new releases to ensure they meet safety standards, The New York Times reported.