White House Asks OpenAI to Delay GPT-5.6 Launch, Citing Security Concerns Over Unprecedented Capabilities
The Trump administration has asked OpenAI to restrict the release of its next AI model, GPT-5.6, to a limited set of government-approved partners before any wider public rollout — marking the first time the U.S. government has preemptively intervened to limit the launch of a domestically developed AI model.
The request came from the White House’s Office of the National Cyber Director and the Office of Science and Technology Policy, according to a source familiar with the matter. The administration is currently building a framework to test and evaluate the security of new AI models before they reach the public.
What Triggered the Intervention
According to a source familiar with the situation, the government stepped in specifically because GPT-5.6 has capabilities described as “Mythos-like” — a reference to Anthropic’s frontier model, which the Commerce Department recently directed the company to revoke access to. Officials are not, the source stressed, signaling a broader shift toward heavy-handed regulation of the AI industry.
“This is what’s happening with models of that caliber,” the source told Axios. “The models are so powerful that the administration wants to be sure the companies have adequate safeguards in place.”
OpenAI CEO Sam Altman disclosed the limited-rollout plans in a memo to employees, first reported by The Information. Altman made clear that OpenAI is not enthusiastic about the arrangement. “We’ve made clear to the U.S. government that this is not our preferred long-term model, and will work with them and others in industry to achieve a more sustainable approach for future releases,” he wrote.
Government Coordination Behind the Scenes
Altman met with Commerce Secretary Howard Lutnick on Wednesday to discuss GPT-5.6, Axios has learned. Lutnick pushed to ensure that all relevant government agencies had the opportunity to test and approve the model before its release.
The White House has already been briefed on the model’s capabilities and given access to preview its functions — an unusual degree of government involvement in a private company’s product launch.
OpenAI, notably, had been working proactively with the administration on the rollout even before Anthropic was directed to revoke access to its own frontier models, Fable 5 and Mythos 5, under a rare Commerce Department directive.
The Regulatory Context
President Trump signed an AI security executive order earlier this month directing several federal agencies to establish a voluntary testing protocol for AI companies ahead of major model releases. The order was delayed for weeks due to internal political disputes over how restrictive and mandatory the program should be.
The administration’s intervention with OpenAI represents the first concrete application of that emerging framework — and a test of how far government oversight of frontier AI will extend.
A Sector Under Pressure
The episode highlights the difficult position AI companies now occupy. Labs like OpenAI are racing to release new models to stay competitive — not only with each other, but with increasingly capable Chinese open-source models that operate outside U.S. regulatory reach.
At the same time, security officials and corporate leaders are raising alarms about what happens when highly capable AI systems fall into the hands of nation-state spies, cybercriminals, or rogue insiders.
The tension between speed-to-market and responsible deployment is not new — but the scale of capability now at stake has pushed that tension into the White House itself.
What Comes Next
Altman indicated in his memo that he hopes to release GPT-5.6 just “a couple of weeks later” than originally planned, suggesting the delay will be short-lived. Whether the government’s voluntary testing framework can keep pace with the industry’s release cycles — or whether future interventions become more frequent and more mandatory — remains an open question.