OpenAI will delay the broad release of GPT-5.6 and instead offer a limited enterprise preview, chief executive Sam Altman told employees Wednesday, complying with a request from the Trump administration over security concerns about the company's next frontier model. Initial access will go to roughly 20 partners through Amazon Bedrock, with federal officials approving additional customers individually during the preview period.

The arrangement differs sharply from treatment of rival Anthropic, which this month suspended global access to Mythos 5 and Fable 5 under an export control directive barring foreign nationals from using the models. Industry executives said the uneven approach raises questions about whether Washington is building predictable rules or negotiating company by company as capabilities advance.

What OpenAI Disclosed

Altman said in an internal question-and-answer session that the government has grown more anxious about cutting-edge model capabilities, according to a person familiar with the meeting. He urged staff to work with administration officials on safety input even when the company disagrees, the person said. OpenAI did not publish a formal statement detailing timelines or evaluation criteria.

Enterprise customers awaiting GPT-5.6 had expected a wider rollout this summer. Several Fortune 500 technology officers said procurement teams paused internal pilots until access rules clarify whether subsidiaries outside the United States may use the model at all.

Policy Context

The request follows an executive order signed earlier in June giving the administration and AI companies 60 days to develop a voluntary framework for pre-release government access to frontier systems. That directive envisioned up to 30 days of federal evaluation before public deployment. OpenAI's staggered preview appears to operationalize that concept without a full pause.

White House officials have publicly emphasized a "speed wins" posture on American AI exports while simultaneously intervening when models demonstrate jailbreak risks or foreign access they consider unacceptable. Anthropic's broader shutdown came after officials concluded Fable 5 guardrails could be bypassed for cybersecurity tasks.

Industry Reaction

Microsoft and Amazon, which distribute OpenAI models to cloud customers, said they are coordinating with the company on Bedrock availability. Google and Meta executives privately told investors that disparate treatment could push labs to lobby for uniform standards rather than ad hoc arrangements.

Venture investors backing enterprise AI startups said delayed access may slow competitive benchmarking against GPT-5.6 but reduces near-term regulatory surprise for the largest distributor. Smaller labs without government relationships worry they lack leverage to negotiate comparable terms.

Competitive Implications

Anthropic continues talks with federal officials to restore Mythos and Fable access under revised controls. OpenAI's limited preview keeps its flagship roadmap visible while Anthropic's top models remain offline for most users — a commercial gap both companies are reluctant to discuss publicly.

Foreign allies, including officials in Tokyo and Brussels, asked whether U.S. customer-by-customer approval will delay NATO-member deployments of American frontier systems. The State Department has not issued public guidance.

What Comes Next

OpenAI has not set a date for general availability. Congressional staff tracking AI policy expect oversight hearings after the July Fourth recess, with members likely to compare OpenAI's preview structure with Anthropic's export directive. For enterprise buyers, the immediate question is whether their organizations qualify for the first 20 partner slots — and how long federal review adds to procurement cycles already measured in quarters.