OpenAI and Anthropic limit new AI models to government-approved users amid federal review
OpenAI and Anthropic said they will restrict access to their latest AI models to a small group of customers cleared by the Trump administration, citing cybersecurity concerns and a temporary government vetting process.
OpenAI announced on Friday that its newest model, GPT-5.6 Sol, will be offered only to a limited set of partners approved by the Trump administration. The company said the restriction is a temporary measure while it conducts a "phased release" with stronger safeguards, and it does not expect the government-access process to become a permanent default. Anthropic followed hours later, confirming that its most advanced cybersecurity model, Mythos 5, has been cleared for a limited rollout to a small group of cyber-defenders and infrastructure providers after a prior ban on foreign-national use. Both firms said the models will remain unavailable to the broader public until further review.
The White House said it continues to work with frontier AI labs to address scaling challenges, while the administration’s AI oversight order, signed by President Trump in June, allows up to a 30-day federal vetting of advanced systems before public release. Critics, including Rep. Lori Trahan, warned that the ad-hoc approval process lacks transparent legal oversight. Stanford cybersecurity expert Alex Stamos said he found no factual basis for the government’s actions, noting that the risks cited are similar to those of other publicly available AI tools.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have been in recent talks with Commerce Secretary Howard Lutnick and other officials as both companies prepare for potential public offerings. Anthropic expressed satisfaction with the partial release of Mythos 5 and indicated it will continue cooperating with the government to restore broader access to its Fable model.