The Trump administration has asked OpenAI to stagger the rollout of its upcoming GPT-5.6 model, requiring government approval before the company can grant access to additional users. Instead of a broad public release, GPT-5.6 will launch as a limited preview available only to a small group of trusted enterprise partners, with the wider public release potentially delayed until 2027.

Why Is the US Government Slowing Down AI Model Releases?

The request comes from two federal offices: the Office of the National Cyber Director and the Office of Science and Technology Policy, both citing national security concerns. The move reflects growing anxiety in Washington about the capabilities of advanced artificial intelligence systems and their potential implications if distributed without safeguards. This represents a significant shift in how the government approaches AI regulation, moving beyond semiconductor export controls to directly managing the deployment of frontier AI models themselves.

OpenAI CEO Sam Altman informed employees during an internal meeting that the company would comply with the staggered rollout request, though he made clear this is not OpenAI's preferred long-term approach. "We've made clear to the US 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 stated.

How Will the Limited Preview Process Work?

The rollout structure introduces a new approval mechanism for frontier AI access. Here's what the phased approach entails:

• Initial Access: OpenAI will provide GPT-5.6 to approximately 20 trusted enterprise partners during the preview phase, with Amazon Bedrock expected to serve as one of the primary platforms for accessing the model.
• Government Approval: The Trump administration will review and approve customer access on a case-by-case basis, meaning organizations cannot simply request access and receive it immediately.
• Delayed Public Release: The broader public launch may be postponed until 2027, though OpenAI had originally planned to release GPT-5.6 within weeks.
• Compliance Requirements: OpenAI employees have been instructed to work with the Trump administration on any input regarding safety and restrictions for the company's upcoming models, even if the company disagrees with the guidance.

The arrangement is more permissive than recent restrictions applied to rival AI company Anthropic, whose Mythos 5 and Fable 5 models were subject to limitations on foreign access after the government cited security concerns. Anthropic had suspended access to these advanced models for foreign users after receiving export control directives from the US government.

What Does This Mean for OpenAI's IPO Plans?

The timing of the GPT-5.6 delay intersects with OpenAI's confidential filing for a US initial public offering earlier this month. The company is reportedly targeting a valuation of up to $1 trillion, though advisers have discussed postponing the listing until 2027 to better support that valuation. Altman has reportedly opposed lowering the company's valuation target, which could create tension if the delayed model release affects investor confidence in OpenAI's growth trajectory.

The government's involvement in model deployment also raises questions about how future AI releases will be managed. Currently, AI companies lack a standardized process for informing US authorities about frontier model launches. There is uncertainty across the industry about which government officials should be briefed, what technical information should be shared, and how far in advance officials should be notified.

Is a Formal AI Approval Framework Coming?

President Donald Trump signed an executive order earlier this month directing the administration and AI companies to develop a voluntary framework for frontier AI models within 60 days. This proposed framework would allow the US government to review certain advanced AI models for up to 30 days before their planned public release, creating a more structured approval process for future launches. The framework could eventually replace the ad hoc case-by-case approvals currently being applied to GPT-5.6.

OpenAI employees are concerned about whether the government's actions against Anthropic could affect the company's ability to launch future AI models on schedule. Anthropic disputed the government's reasoning, arguing that the discovery of a narrow potential jailbreak in Fable 5 should not justify recalling a commercial model deployed to hundreds of millions of people. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers," Anthropic stated in a blog post.

The broader pattern suggests that US policymakers are now paying closer attention to the deployment of cutting-edge AI models as governments seek to balance technological leadership with national security concerns. Rather than focusing only on controlling the hardware and semiconductors used to build AI systems, officials are increasingly focused on controlling who gets access to the most capable models themselves.