whitehouse.gov web signal

Trump EO bans mandatory AI preclearance, sets classified benchmarks

NSA Treasury White_House executive_order ai_policy frontier_models national_security ai_regulation

Key insights

  • The order explicitly prohibits mandatory licensing, preclearance, or permitting requirements for AI model deployment.
  • Treasury and the Department of War have 60 days to build a classified benchmarking system defining what qualifies as a 'covered frontier model.'
  • Treasury has 30 days to establish an AI cybersecurity clearinghouse coordinating vulnerability scanning and remediation across federal systems.

Why this matters

The explicit prohibition on mandatory preclearance removes a regulatory overhang that had been debated across prior administrations, giving frontier AI developers a clearer runway without formal federal gatekeeping. The classified benchmarking regime simultaneously creates an opaque evaluation layer: Treasury and the Department of War will define the covered frontier model threshold in secret, meaning developers cannot know in advance what capability level draws federal scrutiny. For AI practitioners and founders, the voluntary early-access framework carries implicit stakes -- participation may become a de facto prerequisite for federal procurement even without legal compulsion, creating soft pressure the order does not formally acknowledge.

Summary

Trump's June 2 executive order settles one contested question: no mandatory licensing or preclearance for AI models. Developers can engage federal agencies through a voluntary framework to determine whether their models meet the 'covered frontier model' threshold, but nothing can compel them through a government gate before deployment. The binding obligations run to agencies, not developers. Treasury has 30 days to form an AI cybersecurity clearinghouse, in consultation with the National Cyber Director, to coordinate vulnerability scanning and remediation. Treasury and the Department of War jointly have 60 days to build a classified benchmarking process assessing the 'advanced cyber capabilities of AI models' and fix the threshold at which a model earns the covered frontier model designation. Essentially: (Treasury, Department of War) hold the evaluation authority while developers engage voluntarily. - OPM has 60 days to expand the US Tech Force Information Cybersecurity Specialist program. - The Attorney General must prioritize enforcement of computer fraud and wire fraud statutes against criminal AI use. Because the benchmarking is classified, the capability threshold that triggers federal scrutiny will be set outside public view.

Potential risks and opportunities

Risks

  • Frontier AI developers could be designated 'covered frontier models' through a classified process with no public criteria, triggering compliance obligations they had no ability to anticipate or contest.
  • If Treasury's 30-day clearinghouse deadline slips -- the order is explicitly subject to appropriations availability -- federal civilian systems remain uncoordinated against AI-enabled attacks during the gap.
  • The Department of War's role in classified model benchmarking gives the military direct evaluation authority over commercial AI without public accountability, creating a channel through which national security concerns could shape private-sector development invisibly.

Opportunities

  • Federal cybersecurity contractors with existing clearances are positioned to supply the classified benchmarking infrastructure Treasury and the Department of War must stand up within 60 days.
  • Vulnerability scanning and remediation vendors can compete for tooling contracts inside Treasury's AI cybersecurity clearinghouse, which carries a hard 30-day formation deadline and a mandate from the National Cyber Director.
  • OPM's 60-day Tech Force Information Cybersecurity Specialist expansion creates near-term government demand for cleared AI talent, benefiting technical staffing firms with existing security clearance pipelines.

What we don't know yet

  • What specific 'advanced cyber capabilities' will define the covered frontier model threshold -- the EO mandates the benchmarking process be classified, leaving developers with no visibility into the criteria.
  • Whether voluntary early-access participation will become an informal prerequisite for federal contracts despite the explicit prohibition on mandatory preclearance.
  • How Treasury's 30-day AI cybersecurity clearinghouse will coordinate with CISA's separate 30-day directive for civilian federal systems, and which agency holds final authority over remediation priorities.

Originally reported by whitehouse.gov

Read the original article →

Original headline: Trump Signs Executive Order Promoting Advanced AI Innovation and Security