The Trump administration has formally asked OpenAI to delay the public release of its newest flagship model, GPT-5.6, according to multiple reports from Washington. The request, confirmed by officials familiar with the conversations, marks the most direct intervention by a sitting US government into a frontier artificial intelligence launch and signals a sharp escalation in the political fight over how rapidly the most capable AI systems should reach the public.
Politico first reported on Thursday evening that senior White House staff had reached out to OpenAI leadership, urging the company to hold the GPT-5.6 rollout until federal reviewers could complete an additional round of safety evaluations. The intervention came just hours after OpenAI executives had privately briefed journalists that the new model was ready for a wide release window beginning in early July. That timeline now appears to be in question.
Inside the administration, the push for delay is being driven by a coalition of officials who have grown increasingly vocal about the pace at which frontier labs are shipping systems with capabilities that have not been fully stress-tested. According to people familiar with the discussions, the administration’s concerns center on three areas: the model’s reported improvements in agentic task completion, its expanded context window, and its ability to autonomously execute multi-step operations across consumer software. Each of those capabilities has been a focus of internal policy debates for months.
Why the administration moved now
For most of 2026, federal AI policy has been shaped by the voluntary commitments major labs made through the White House AI Safety Institute and through successive executive orders on safe, secure, and trustworthy AI development. Those commitments, while real, did not give the executive branch any formal authority to block a commercial product launch. Officials making the request to OpenAI have emphasized that the conversation is advisory rather than mandatory, but they have also made clear that continued cooperation on future federal procurement decisions may depend on how the company responds.
The move lands at a sensitive moment for OpenAI. The company has been working to keep its enterprise momentum following last year’s corporate restructuring, and a delayed flagship launch would put it further behind competitors that have moved faster on agentic systems. Anthropic and Google DeepMind have both shipped widely available model upgrades in recent weeks without triggering the kind of high-level scrutiny now focused on OpenAI. The asymmetry of the response has prompted speculation inside the AI policy community that the delay request is at least partly political, though administration officials have publicly framed it as a straightforward safety question.
The capabilities that triggered the review
- Agentic execution: GPT-5.6 is reported to handle multi-step tasks across productivity software with significantly less human oversight than its predecessor.
- Long-context reasoning: The model is said to retain and act on information across extremely long documents, raising concerns about potential misuse for large-scale analysis of sensitive data.
- Autonomous tool use: Internal benchmarks leaked earlier this month suggested GPT-5.6 can chain dozens of tool calls without intermediate review.
“The conversation is not about whether GPT-5.6 is safe to ship. It is about whether the company has demonstrated that its evaluation process is rigorous enough to deserve the public’s trust,” said one congressional staffer familiar with the discussions.
Industry reaction
Reaction across the AI industry has been sharply divided. Investor groups and a number of enterprise customers have pressed OpenAI not to alter its release plans, arguing that any sign of political intervention in commercial AI deployment would damage the United States’ competitive position against Chinese labs. Other safety-focused researchers have welcomed the administration’s caution, noting that recent frontier model evaluations have surfaced capability jumps that even the labs themselves did not fully anticipate.
OpenAI has not publicly committed to a specific response, but people familiar with the company’s posture say executives are weighing a middle path: a limited release to enterprise customers under additional safety guardrails, paired with a delay of the consumer-facing launch until certain evaluation milestones are met. That approach would allow OpenAI to preserve its enterprise revenue stream while giving the administration a tangible concession.
What happens next
For now, GPT-5.6’s launch date remains officially unchanged, but the practical timeline is now in flux. The next inflection point is likely to come within the next two weeks, when OpenAI is expected to issue a formal update on its release plans. In the meantime, the episode is likely to accelerate a broader debate in Washington about whether the United States needs a more explicit statutory framework for frontier AI deployment — one that gives the federal government clear authority to act on safety concerns rather than relying on voluntary cooperation from the companies developing the technology.
The international picture
Beyond Washington, AI regulators in Brussels and London are watching the OpenAI situation closely, as their own frontier AI frameworks are still in early implementation. The European AI Office, which has been coordinating the rollout of the EU AI Act’s general-purpose model obligations, has not signaled any direct response to the US delay request, but officials there have noted in recent weeks that they expect closer cooperation with Washington on frontier model evaluations going forward. For OpenAI and its peers, the practical implication is that a single regulatory conversation in Washington now has the potential to ripple across multiple jurisdictions — a coordination challenge that frontier labs have been quietly preparing for since the start of the year.