The White House wants to vet AI before you can use it

The White House has shifted its stance on artificial intelligence, issuing an executive order under President Donald Trump that directs the most powerful AI companies to share their newest models with the government for review before public release. The order, titled 'Promoting Advanced Artificial Intelligence Innovation and Security,' introduces a voluntary framework where labs can provide early access to 'covered' frontier models for up to 30 days prior to launch, though no model is required to pass a review before shipping. The initiative follows concerns sparked by Anthropic’s Mythos Preview, a frontier AI model that autonomously identified over 23,000 software vulnerabilities in widely used systems, including more than 6,000 rated high or critical severity. Roughly 91% of these vulnerabilities were later validated by independent firms, raising alarms about the potential risks of unchecked AI capabilities. Anthropic initially granted early access to about 50 partners through Project Glasswing before expanding it to 150 organizations across 15 countries in early June. Beyond model review, the executive order emphasizes cybersecurity, directing federal agencies to fortify systems tied to national security, the Department of War, and civilian government operations. The framework also aims to expand cyber hiring to bolster defenses. While participation remains voluntary, the order leverages classified benchmarks from the National Security Agency (NSA) and a government-managed review process to quietly establish a permanent vetting mechanism for advanced AI. The White House’s move marks a departure from its previous hands-off approach, driven by fears of ceding technological leadership to China rather than concerns over AI risks. The order reflects growing institutional pressure to monitor AI development before potential threats materialize, though it stops short of mandatory oversight. Experts suggest the framework may eventually serve as a model for broader regulation, with future AI deployments—such as those in workplace tools or consumer apps—potentially subject to federal review before public release.