President Trump quietly signed an executive order Tuesday that requires AI companies to submit powerful new models for voluntary government review 30 days before public release. The directive aims to give federal agencies time to assess potential threats to financial systems, national security and critical infrastructure. This represents a significant rollback from the original proposal that would have mandated 90-day reviews.

Industry pressure drove the dramatic shift. Top White House officials had already approved a 90-day version, with OpenAI, Anthropic and Google reviewing the draft. Former AI czar David Sacks warned the longer timeline would hamper innovation and slow U.S. competition with China. Trump abruptly canceled the May 21 signing ceremony hours before it was set to begin, citing concerns about regulatory overreach.

The order establishes a cybersecurity clearinghouse within 30 days and creates classified benchmarking to evaluate national security risks. It directs the Pentagon to secure networks and instructs the Justice Department to prosecute AI-assisted hacking. These provisions mirror earlier drafts despite the shortened review period.

Former Trump adviser Dean Ball criticized the policy, questioning what intelligence agencies could accomplish in 30 days to improve safety. The order emerges amid mounting concerns about models like Anthropic's Mythos, which researchers say has already discovered buried vulnerabilities in critical systems. The voluntary approach signals continued tension between security imperatives and innovation goals.