The US government has partially reversed its June 12 order that forced Anthropic to disable its most powerful artificial intelligence models, while simultaneously requiring rival OpenAI to restrict the launch of its new GPT-5.6 series to a government-vetted group of partners. The Commerce Department informed Anthropic on Friday that it may now supply its Mythos 5 model to more than 100 US companies and institutions, including Fortune 500 firms, after the company addressed security risks identified earlier this month. The less powerful Fable 5 model remains suspended. Hours later, OpenAI began a limited preview of its Sol, Terra, and Luna models, with access granted only to a small set of trusted partners whose identities were shared with the administration.

The twin moves crystallise an ad hoc federal gatekeeping mechanism for frontier AI systems whose advanced cybersecurity capabilities have alarmed national security officials. Both Mythos 5 and GPT-5.6 Sol can autonomously identify and chain software vulnerabilities at a speed and scale that, in the wrong hands, could be weaponised against critical infrastructure. President Trump signed an executive order earlier in June establishing a voluntary 30-day pre-release review process, but the framework remains undefined. In practice, the government is now approving access client by client, a process that OpenAI’s chief executive Sam Altman told staff was not the company’s preferred long-term approach.

The partial easing for Anthropic does not resolve the deeper friction between the safety-focused company and the administration. The Pentagon earlier designated Anthropic a supply-chain risk after it refused to permit use of its technology for mass surveillance and autonomous weapons, a dispute now being litigated in two lawsuits. Viewed from Washington, the differential treatment of the two leading AI labs has drawn accusations of political favouritism, though officials insist the interventions are driven solely by security assessments. In Congress, Democratic Representative Lori Trahan warned that the administration is “deciding company by company who gets access to the newest AI model” with no statutory process or oversight.

Anthropic said it is working to restore access to Fable 5 and expand the Mythos 5 rollout, while OpenAI expects broader availability of GPT-5.6 within weeks. The executive order gives the administration 60 days to develop a permanent voluntary framework with industry, a deadline that will test whether the current improvisation can be replaced by a predictable regulatory architecture before the next generation of models arrives.