Trump executive order mandates pre-release cybersecurity testing for AI models
Anthropic's Mythos Preview found thousands of critical vulnerabilities, sparking new federal oversight.
President Donald Trump signed a new executive order directing leading AI companies to voluntarily submit their flagship models for government cybersecurity testing before market deployment. The order, reported by Reuters, tasks the Departments of Treasury, Defense, Commerce, and Homeland Security with securing agreements with AI developers to test advanced cyber capabilities of models. Agencies will have up to one month to review models before public release.
The move was spurred by Anthropic's release of Mythos Preview, an AI model reportedly capable of identifying decades-old software vulnerabilities and developing working exploits. Though not publicly available, the tool has already uncovered "thousands" of vulnerabilities, including critical-severity ones. Major AI developers expressed support: Google's Kent Walker called it an important step, Anthropic said it looks forward to working with the White House, and OpenAI CEO Sam Altman noted the order "gets the balance right." The order reflects a bipartisan convergence on treating frontier AI as a strategic capability similar to dual-use military technologies.
- Trump's EO requires voluntary pre-release cybersecurity testing of frontier AI models by government agencies.
- Anthropic's Mythos Preview, an AI vulnerability finder, discovered thousands of critical-severity software flaws, prompting the order.
- Major AI firms (OpenAI, Google, Anthropic) support the initiative; xAI and Microsoft also recently agreed to federal testing.
Why It Matters
Frontier AI is now treated as a strategic national security asset, shifting from hands-off to proactive government oversight.