Trump admin to require 90-day safety review for AI models like Anthropic's Mythos
New executive order gives feds 90 days to vet AI vulnerabilities before public release.
The Trump administration is poised to sign an executive order establishing a voluntary framework for federal agencies to audit powerful new AI models—such as Anthropic's Mythos—before they reach the public. Under the plan, developers would submit models to agencies like the Department of Homeland Security up to 90 days ahead of release to identify safety vulnerabilities, similar to the voluntary commitments seen under the Biden administration's AI executive order. The move aims to preempt catastrophic risks while avoiding heavy-handed regulation that could stifle innovation.
Simultaneously, the U.S. Export-Import Bank (EXIM) is expected to approve a program providing billions in financing to foreign buyers of American AI products, including software, tools, and advanced chips. However, purchases of sensitive technologies like high-performance semiconductors would require sign-off from the Commerce Department. This dual strategy seeks to keep the U.S. competitive globally—especially against China—by both tightening domestic oversight and boosting American AI exports.
- Voluntary federal review: AI models (e.g., Anthropic's Mythos) can be vetted up to 90 days pre-release for vulnerabilities.
- EXIM financing program: Billions in loans for foreign purchases of U.S. AI tools, with Commerce oversight for sensitive chips.
- Balances safety and competitiveness: Aims to prevent catastrophic AI risks while accelerating American AI exports abroad.
Why It Matters
Balances AI safety oversight with aggressive export financing, shaping U.S. leadership in global AI race.