Trump Readies AI Order With 60-Day Benchmarking and 90-Day Model Review

7 articles · Updated · POLITICO · May 20

A Trump executive order could arrive as soon as Thursday, directing federal agencies to expand AI hiring and build a classified process for identifying which systems qualify as covered frontier models.
OPM would boost recruiting for the U.S. Tech Force, while Treasury, CISA and NIST would get 60 days to design the benchmarking framework; the NSA would make final determinations with other agencies.
AI developers joining the voluntary framework would be asked to notify the government before releasing covered models, provide access 90 days before public launch, and share access to select critical infrastructure ahead of rollout.
The push follows alarm over cyber-focused models such as Mythos, whose limited debut last month triggered White House talks with tech and cyber firms even as internal disagreements emerged over how tightly to police frontier AI.

As America tightens AI review, what stops unregulated foreign models from dominating the global landscape?

Can a 90-day voluntary review truly safeguard against AI models that can generate novel threats instantly?

When a government-vetted AI causes catastrophic harm, who will ultimately be held accountable?