The US government has asked OpenAI to hold back its newest AI model, GPT-5.6, releasing it only to a small group of partners that Washington signs off on first, according to a report by The Information. The move comes barely two weeks after the same administration forced Anthropic to yank its Fable 5 and Mythos 5 models offline entirely, signaling that the White House is now actively policing how the most powerful AI systems reach the public. CEO Sam Altman told staff the controlled rollout wasn’t OpenAI’s “preferred” approach, but the company agreed to it anyway, with federal officials set to approve customer access one by one during the preview window.The request came from the White House’s Office of the National Cyber Director and the Office of Science and Technology Policy, with Commerce Secretary Howard Lutnick also in the loop. Altman reportedly discussed GPT-5.6 directly with Lutnick on Wednesday, and the Commerce chief wanted assurance that every relevant arm of the government had tested and cleared the model before any launch. The Information first surfaced the plan, citing a leaked internal memo Altman sent to employees.
A softer landing than the one Anthropic got
What stands out is how differently the two companies have been treated. When Anthropic shipped Fable 5 on June 9, it took just three days for the government to step in. By the evening of June 12, Commerce had issued an export control directive barring any foreign national, including Anthropic’s own non-citizen employees, from touching Fable 5 or Mythos 5. The net was so wide that Anthropic had no choice but to switch off both models for everyone. The company was reportedly handed a 90-minute window to comply.OpenAI, by contrast, gets to keep its model available, just to a vetted list. Sources told Axios the company had been working with the administration on the release well before the Anthropic blowup, which likely smoothed the path. One source framed the intervention plainly: GPT-5.6 carries “Mythos-like” capability, and that level of power is what triggered the extra scrutiny rather than any sudden policy shift.
Washington’s new rulebook for AI is still being written
All of this traces back to an executive order Trump signed on June 2, which asks AI labs to give the government up to 30 days to evaluate powerful new models before release. The catch is that the order is officially voluntary and the actual review framework doesn’t have to be finished until the end of July. So companies are navigating a process whose rules keep shifting and, by some accounts, aren’t fully settled even inside the administration. Dean Ball, a former Trump AI policy adviser who recently joined OpenAI, captured the mood bluntly, noting that the requirements change constantly and stay secret even to the officials applying them.That uncertainty has spread across the industry. The government is separately pressing Meta, the lone major holdout, to submit its models for review. OpenAI, Anthropic, Google, xAI and Microsoft have all already agreed to share their systems with the Center for AI Standards and Innovation, the Commerce-housed group Lutnick oversees.
Why US government suddenly cares so much about AI launches
The deeper worry driving all this is cybersecurity. Officials are increasingly anxious about what happens when nation-state hackers, cybercriminals or rogue insiders get hold of frontier models capable of finding and exploiting software flaws. That same fear is what sank Fable 5, after Amazon researchers showed they could coax the model into revealing security vulnerabilities. Critics argued the panic was overblown, since rival models, including OpenAI’s own, can do much the same.For now, Altman says he hopes to push GPT-5.6 to a wider audience just a “couple of weeks” after the limited preview. Whether that timeline holds depends on a review process the government is still building in real time.
