The Trump administration has pressured OpenAI to delay and restrict the rollout of GPT-5.6, its next flagship model, requiring the company to limit initial access to a vetted group of government-approved partners before any public launch. The move marks the first time Washington has preemptively blocked a major American AI company from releasing a model on its own timeline.
The White House's National Cyber Director and Office of Science and Technology Policy made the request as the administration builds out formal security protocols for evaluating new AI systems before they reach the broader market. OpenAI CEO Sam Altman confirmed in a memo to staff that the company accepted the restricted rollout approach, though he signaled it was a temporary 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," Altman wrote in the memo.
Commerce Secretary Howard Lutnick spoke directly with Altman on Wednesday to ensure all relevant agencies could test and sign off on the model's safety before any broader availability. A source said the intervention centered on GPT-5.6's exceptional capabilities, which officials described as comparable to systems that prompted similar government action from the Commerce Department in recent weeks.
The government is not taking a sudden hard line, according to insiders familiar with the decision. Rather, officials concluded that models of this power level require adequate safeguards and third-party validation before entering the wider market, where bad actors, foreign intelligence services, and cybercriminals could potentially gain access.
OpenAI has been collaborating with the White House since before the Commerce Department directed Anthropic to restrict access to its frontier models. Government officials have already had a chance to preview GPT-5.6's abilities and have been looped in on technical details throughout development.
Altman told staff he expects to move forward with the wider release "a couple of weeks later" once the government's testing concludes, suggesting the delay may be measured in weeks rather than months.
The order sits within a broader executive action on AI security that President Trump signed earlier this month. That order directs federal agencies to establish a voluntary testing protocol for AI companies, though political disagreement over whether such programs should be mandatory delayed the directive for weeks before it advanced.
AI companies face mounting pressure from multiple directions. They must accelerate releases to compete with one another, yet they also confront a surging wave of open-source models from China that grow more capable monthly. Now they must navigate a government apparatus that wants visibility into, and control over, how the most powerful systems reach the public.
Security leaders and corporate executives have grown increasingly vocal about the risks of uncontrolled access to frontier AI. Nation-state actors, criminal networks, and insider threats all represent plausible vectors for misuse if these models spread without proper gates.
Author James Rodriguez: "This is the shape of AI regulation going forward: not bans, but gatekeeping. OpenAI caved faster than anyone expected, which suggests the industry may be more compliant than recent posturing suggested."
Comments