Trump administration tightens review of new AI models, limits public access

OpenAI said it was restricting access to its newest model, GPT-5.6 Sol, to customers approved by the government, a sharp sign that the Trump administration is using national-security review to shape who can see frontier AI first. Anthropic moved in the same direction hours later, saying the government had approved only a limited release of its strongest cybersecurity model while vetting continued.

The new framework, created after Donald Trump signed an executive order on AI oversight earlier in June, allows the government to hold advanced models for up to 30 days before public release. OpenAI said the limited rollout came at the administration’s request and described it as temporary, with broader availability planned in the coming weeks. But the effect is immediate: instead of a public launch, the first users are being narrowed to small groups of trusted partners and other customers acceptable under federal review.

Anthropic’s case shows how quickly that gatekeeping can turn into an access filter. On June 12, the company said, the U.S. government issued an export-control directive at 5:21 p.m. ET ordering it to suspend access to Fable 5 and Mythos 5 for any foreign national, including its own foreign national employees. Anthropic said Mythos 5 would initially be deployed through Project Glasswing with the U.S. government, as an upgrade to Claude Mythos Preview, and that a small group of cyberdefenders and infrastructure providers would get access first.

The company has framed that caution around concrete security risks. Anthropic said it and its approximately 50 Project Glasswing partners used Claude Mythos Preview to find more than 10,000 high- or critical-severity vulnerabilities, with some partners seeing bug-finding rates rise by more than a factor of ten. It also said it spent thousands of hours red-teaming Fable 5 with the U.S. government, the U.K. AISI, private third parties and internal teams before launch. Anthropic priced Fable 5 and Mythos 5 at $10 per million input tokens and $50 per million output tokens, and said its safety filters sometimes catch harmless requests in less than 5% of sessions.

The White House and the Commerce Department are now asserting a practical power over advanced AI distribution that looks less like ordinary product review and more like licensing through procurement and national-security channels. That shift reaches beyond chatbots, touching cybersecurity tools, enterprise deployments and foreign access. Anthropic said in November 2025 that cyber capabilities had doubled in six months and that it had disrupted what it called the first documented large-scale cyberattack executed without substantial human intervention, a warning that helps explain why the government is treating frontier models as dual-use systems before they reach the public.