Key Points
- OpenAI said it was withholding the wide release of its latest AI model, GPT-5.6, at the request of the U.S. government.
- The Trump administration asked OpenAI to begin with a limited preview for a small group of trusted partners before a broader rollout.
- On the same day, the U.S. Commerce Department told Anthropic that export controls on its Mythos 5 model would be relaxed.
- Anthropic said Mythos 5 could be redeployed to a set of U.S. organisations operating and defending critical infrastructure.
- The restoration is expected to cover around 100 organisations, including government agencies and private companies.
- Anthropic still said talks with government officials would continue over the weekend over access to its Fable 5 model.
Washington, D.C. (Evening Washington News) 30 June 2026 — At the heart of Anthropic’s clash with the U.S. government is a choice by the company to resist what one report describes as the new operating rules of Trump’s Washington, while OpenAI moved to comply with a government request to limit access to GPT-5.6.
As reported by Fortune, OpenAI announced on Friday that it would withhold the wide release of GPT-5.6 at the request of the U.S. government, while Anthropic was told by the Commerce Department that export controls on Mythos would be eased after a two-week ban forced the model offline for all users.
The shift marks a striking contrast in how two of the biggest AI firms are dealing with Washington. OpenAI is limiting its rollout to a small group of trusted partners, while Anthropic has been pushing to restore access after a far tighter intervention from the administration.
Why did OpenAI limit GPT-5.6?
OpenAI said its new GPT-5.6 models — Sol, Terra and Luna — were first being given to a small group of trusted partners, with wider availability expected in the coming weeks.
The company said it had previewed the models to the U.S. government before launch and was complying with the request to slow the initial release.
According to the reporting, this was presented as part of a broader national security approach around frontier AI models, with the government seeking more control over who gets early access.
What changed for Anthropic?
Anthropic said on X that the government had notified it that Mythos 5, described by the company as its strongest cybersecurity model, could be redeployed to a set of U.S. organisations that operate and defend critical infrastructure.
A letter seen by NBC News and reported by The Tribune said Commerce Secretary Howard Lutnick informed Anthropic that trusted users could once again access the model.
The restoration is expected to include around 100 organisations, among them government agencies and private companies, according to people familiar with the matter.
How did the ban begin?
Two weeks earlier, the Commerce Department had invoked export control authorities to require Anthropic to suspend access to Mythos 5 and its consumer model, Fable 5, citing national security threats.
The restrictions came after concerns emerged about the models’ cyber capabilities and a reported jailbreak issue connected to Fable, which reportedly raised fears that users could reach more powerful underlying capabilities.
As reported by Fortune, the administration’s actions followed a wider dispute in which Anthropic had refused to accept some of the political and contractual pressures that other firms were more willing to accommodate.
What do the reports say about the company’s position?
The Fortune report says Anthropic has taken a different line from many of its peers by not trying to flatter or appease the White House.
It also says the company has faced repeated criticism from Trump administration figures, including attacks directed at chief executive Dario Amodei, while Anthropic has argued that it should be able to speak openly when it disagrees with government policy.
At the same time, the report says Anthropic has tried to improve its Washington strategy by adding Republican and former Trump administration figures to its policy and federal affairs teams.
What does this mean for the AI race?
The reporting suggests the White House is now influencing how frontier AI models are released, tested and distributed, rather than simply reacting after launch.
That means model rollout may increasingly depend on government approval, trusted-partner access, and negotiations over security concerns.
For AI firms, the immediate effect is a slower and more controlled path to market. For regulators, it creates a way to limit access before a model reaches the wider public.
What did Anthropic say about access restoration?
Anthropic said discussions with government officials would continue over the weekend as it sought to restore access to Fable 5 as well.
The company said Mythos 5 could be used again by U.S. organisations involved in critical infrastructure defence, but the restrictions had not been fully lifted for all products.
That leaves the company in a partial reopening rather than a full return to normal operations.
Background of this development
The broader background is a growing confrontation between the Trump administration and leading AI firms over national security, export controls and who should get access to powerful models first.
OpenAI’s limited release of GPT-5.6 and Anthropic’s earlier shutdown of Mythos and Fable both reflect a new environment in which model launches can be shaped by direct government intervention.
The Fortune report places Anthropic’s conflict in the context of its refusal to follow the usual Washington playbook of political accommodation, which the company appears to have treated as secondary to its own policy principles.
Prediction
For AI developers, this development is likely to mean more staged rollouts, more security reviews and more direct contact with regulators before public launches.
For enterprise users, especially those in cybersecurity and infrastructure, access may become more selective but also more politically and legally structured.
For the wider audience, the immediate effect is likely to be slower access to new AI tools, while the longer-term effect may be a more tightly managed AI market in which government permission matters as much as technical readiness.
