OpenAI’s next generation of AI models, GPT-5.6, is not arriving as a normal public launch. The company confirmed on Friday that it is delaying broad access at the request of Trump’s White House, creating an unusual release process for one of the most closely watched AI model updates.
Instead of opening GPT-5.6 to all users at once, OpenAI said it will first share the models with a small set of customers. Those customers will be pre-approved by the US government. The company then plans to work with the administration to slowly expand availability.
What is changing about the GPT-5.6 release
The central change is not that OpenAI has canceled GPT-5.6. It is that access is being staged through a government approval process before the models become widely available.
According to the source article, OpenAI is not happy about the delay, based on a person familiar with the company’s thinking. Still, the company appears to view the arrangement as temporary. In a blog post, OpenAI said it hopes GPT-5.6 can be made available to everyone in the coming weeks.
The delay was first reported by The Information. OpenAI has since confirmed the basic shape of the release: limited access first, approved customers first, and broader rollout later.
OpenAI’s own blog post framed the decision as a short-term move, not a preferred future standard. The company wrote: “We don’t believe this kind of government access process should become the long-term default.” It also said the process can keep advanced tools away from users, developers, enterprises, cyber defenders, and global partners who need them.
That tension is the core of the story. OpenAI wants to release a powerful new model family, but it is doing so while the White House is still shaping how it wants frontier AI labs to handle cybersecurity concerns.
Why the White House is involved
The White House involvement is tied to cybersecurity concerns around powerful new AI models. Earlier this month, President Trump signed an executive order aimed at those concerns. The order said the White House would create a “voluntary process” for AI labs to share their models with the government 30 days before a broader release.
The order also included a carveout. It said the US government would not turn the voluntary process into a de-facto “licensing regime.” But OpenAI executives said in a Friday briefing that no such voluntary framework exists yet.
That leaves frontier AI labs in an awkward interim period. The source article describes the current environment as one where working with the US government on an AI model launch does not seem especially voluntary, even though the intended framework is supposed to be voluntary.
For OpenAI, the result is a launch that depends on a review path the company cannot fully explain publicly. Executives said OpenAI sends the US government a list of customers and then gets feedback on it. They also said they cannot share details about exactly how the White House is approving those customers.
The Anthropic case adds pressure
OpenAI is not the only AI lab affected by the administration’s recent moves. The White House asked OpenAI to stagger its release just two weeks after it sent an export control directive to Anthropic.
That directive prompted Anthropic to take its most advanced AI models offline for all customers. Anthropic’s dispute with the White House is still unresolved, according to the source article, and some of Anthropic’s own employees are still barred from using its most advanced AI models.
Together, the OpenAI and Anthropic cases create uncertainty for other US AI labs. Over the last two years, the Trump administration has tried to clear regulation and red tape that could hinder America’s AI innovation and potentially hurt competitiveness with China. More recently, the White House has become increasingly concerned about the cybersecurity abilities of new AI models and has moved quickly to address the issue.
That creates a difficult balance. On one side is the push to move fast and keep US AI development competitive. On the other side is concern that the most capable models may have cybersecurity abilities that deserve closer attention before broad release.
Who gets access first
OpenAI plans to broaden the set of customers that can use GPT-5.6 next week. That expanded group is expected to include some international partners.
For now, access depends on government review. OpenAI has said only that it submits a list of customers to the US government and receives feedback. The White House did not immediately respond to WIRED’s request for comment.
The limited-access phase matters because GPT-5.6 is not a single model release. OpenAI says the family will come in three versions:
- Sol, the most capable version of the model.
- Terra, a middle-tier version of the model.
- Luna, a fast and affordable version.
OpenAI says GPT-5.6 Sol is its most capable model yet on benchmarks testing cybersecurity, biology, and agentic abilities. The company also says it has a “layered safeguard stack” designed to stop bad actors from using its AI model for cyberattacks and other malicious behaviors.
What this means for AI releases
The GPT-5.6 delay shows how frontier AI model launches are becoming more complicated. A release is no longer only a product decision by an AI company. It can also become a government coordination problem, especially when cybersecurity concerns are involved.
OpenAI’s position is that the current approval process should not become the long-term default. At the same time, the company says it is taking the short-term step because it sees it as the strongest path to broader availability in the coming weeks while it works with the Administration on the cyber Executive Order framework and a repeatable process for future model releases.
For users, developers, enterprises, cyber defenders, and global partners, the practical effect is simple: GPT-5.6 exists, but most people cannot use it yet. Access will begin with a small group, expand next week to more customers including some international partners, and then, if OpenAI’s stated hopes hold, reach everyone in the coming weeks.
The unresolved question is whether this becomes a one-time delay during an interim period or a sign of how the most advanced AI models will be released from now on.