Trump signs scaled-back AI executive order

President Donald Trump privately signed a scaled-back executive order on artificial intelligence that establishes a voluntary framework for the federal government to review powerful AI models before public release, Reuters, Politico, The Guardian, and SiliconANGLE report. The final text asks developers to voluntarily provide access to qualifying models for up to 30 days prior to public release, Reuters and The Guardian report. Reporting by Politico and Business Insider says an earlier draft had proposed as much as 90 days of early access; that version was postponed before the final signing. SiliconANGLE reports the order directs agencies to develop a classified benchmarking process and says officials will have 60 days to implement the workflow.

The order uses the term covered frontier model for systems with advanced cybersecurity capabilities, per SiliconANGLE. Reuters and other outlets describe the policy as focused on models with significant potential to affect national security or critical infrastructure. SiliconANGLE highlights Anthropic's preview model Claude Mythos Preview as an example of a system with deep vulnerability-finding capability; SiliconANGLE reports Anthropic said the model has discovered thousands of vulnerabilities in testing. The directive also tasks agencies with creating voluntary testing agreements, expanding AI-powered cybersecurity tools for federal and local users, and instructing the Office of Personnel Management to bolster cybersecurity recruitment, SiliconANGLE reports.

Industry-pattern observations: Voluntary pre-release testing windows are a middle path between mandatory prepublication review and entirely market-driven deployment. For practitioners, such windows can create short operational windows to run focused red-team and vulnerability scans, but they do not mandate specific mitigation outcomes. Similar voluntary schemes typically rely on shared benchmarks and standardized disclosure formats to scale; absent those, ad-hoc reviews can be uneven across vendors and agencies.

Reporting frames the order as a policy shift from an earlier, more deregulatory posture. Reuters notes the move represents a more active White House role on AI security compared with earlier statements from the president. Politico and The Guardian report that technology executives and other advisors lobbied to narrow the original draft, and that the president delayed a previous signing because he said it could undercut U.S. competitiveness with China; Politico and The Guardian both quote the president saying, "I didn't like certain aspects of it. I postponed it." Public coverage characterizes the final order as a compromise: greater federal attention to AI cybersecurity while stopping short of mandatory pre-release controls.

Editorial analysis: Observers should track several measurable indicators over the coming weeks and months.

• •Whether major model developers enter voluntary agreements and the scope of those agreements, as reported by Reuters and Axios.
• •The content and timelines of the classified benchmarking process and the public voluntary framework that agencies are instructed to develop, which SiliconANGLE reports must be implemented within 60 days.
• •How agencies operationalize testing in practice, including what metrics and disclosure formats they request; industry experience shows the choice of metrics determines how broadly testing can scale.

Industry-pattern observations: If major vendors participate, security teams inside enterprises and critical infrastructure operators may benefit from earlier model threat assessments and mitigation guidance. However, because the process is voluntary and time-limited, independent security teams and open-source projects may still need to maintain parallel threat assessments. Finally, the order's emphasis on voluntary review means regulatory outcomes will depend heavily on subsequent agency guidance and any future legislative action.