The US government has inserted itself into who gets to use the most advanced AI models, with both OpenAI and Anthropic agreeing to restrict their latest releases at the Trump administration's request during a cybersecurity review.
According to the Seattle Times and other outlets carrying the wire report, OpenAI said Friday it is limiting the rollout of its new model — identified by BankInfoSecurity as GPT-5.6 — to customers approved by the administration. The government cited cybersecurity risks tied to the models' powerful capabilities.
The sharper episode involves Anthropic. According to India Today, the government two weeks earlier ordered Anthropic to suspend access to some of its models over fears they could be used by military intelligence operators in China or Russia. Anthropic said via its official account that it had been working with the government since June 12 to restore access to its Claude Mythos 5 and Fable 5 models.
That effort has now partially paid off. According to TechCrunch and CNBC, the administration is allowing Anthropic to redeploy Mythos — described by Engadget as a cybersecurity-focused model — to more than 100 trusted US companies and government agencies, including their non-American employees. Anthropic said Mythos 5 is cleared for critical infrastructure operators, while it works to restore general access to Fable 5. The Verge, which reviewed a government letter to Anthropic, described a two-week "rollercoaster" negotiation.
Why it matters: this is an unprecedented move toward government control over which businesses and agencies can use frontier AI, turning powerful commercial models into something closer to export-controlled technology.