White House Prepares AI Security Order Without Mandatory Model Testing
WASHINGTON — The White House is preparing a new executive order on artificial intelligence security that would exclude mandatory testing requirements for AI models, according to a report by Bloomberg.
The order, which is still being finalized, represents a departure from the approach taken by the Biden administration, which had pushed for more prescriptive safety evaluations of frontier AI systems. The decision to omit mandatory model testing requirements reflects the Trump administration’s preference for voluntary rather than mandatory oversight of AI development.
The policy shift directly affects major AI labs operating in the United States, including OpenAI, Anthropic, Google and Meta, which have been developing increasingly powerful AI models amid growing debate over how — and whether — the federal government should mandate safety testing before deployment.
Under President Biden’s October 2023 executive order on AI, companies developing the most powerful models were required to share safety test results with the federal government before releasing those systems to the public. President Trump revoked that order shortly after taking office in January 2025, and the forthcoming order appears to chart a different course on oversight.
Industry groups have argued that mandatory testing requirements could slow American AI innovation and cede competitive ground to China. AI safety researchers and advocacy organizations have called for binding safety evaluations, particularly as AI systems become more capable.
The executive order is expected to focus on security dimensions of AI — including protections against misuse and threats to critical infrastructure — while leaving questions of model safety testing to voluntary industry commitments rather than government mandates.
The policy direction comes as Congress continues to debate multiple legislative proposals addressing AI governance, none of which have advanced to a floor vote in either chamber. Several bipartisan bills introduced in the current session have proposed varying levels of mandatory evaluation for high-risk AI systems, but legislative action remains uncertain.
The order also arrives amid a broader global debate over AI regulation. The European Union’s AI Act, which includes mandatory conformity assessments for high-risk AI systems, is set to take full effect in August 2026. The divergence between U.S. and EU approaches could create compliance challenges for American AI companies operating in both markets.
