OpenAI is delayed the company confirmed on Friday the public release of its next-generation artificial intelligence models, GPT-5.6, at the request of the Trump White House. OpenAI said it would first make the models available to a petite group of customers, which will be pre-approved by the U.S. government. It will then work with the administration to slowly expand access.
According to a person familiar with the company’s thinking, OpenAI is not joyful about this, but believes the delay and government approval process is only short-lived. In a blog post, the company said it hopes to be able to roll out GPT-5.6 to everyone in the coming weeks. The first was OpenAI’s plan to delay the next generation of artificial intelligence models at the request of the Trump administration reported via Information.
“We do not believe that this type of government access process should become the long-term default,” OpenAI wrote in its blog post. “This ensures the best tools are made available to users, developers, enterprises, cyber defense and global partners who need them. We are taking this short-term step because we believe it is the best path to broader availability in the coming weeks as we work with the administration to develop a framework of cyber executive orders and a repeatable process for future model releases.”
Earlier this month, President Trump signed an executive order aimed at addressing the cybersecurity concerns of powerful recent artificial intelligence models. The order said the White House would create a “voluntary process” through which artificial intelligence labs could share their models with the government 30 days before wider publication. The mandate included an exception that stated the U.S. government would not turn its voluntary process into a de facto licensing system for the release of artificial intelligence models. However, during a briefing on Friday, OpenAI executives said no such voluntary framework existed yet. As a result, border AI labs find themselves in a very strange transition period where working with the US government to bring an AI model to market doesn’t seem at all voluntary.
The White House is asking OpenAI to delay the release of artificial intelligence models just two weeks after sending Anthropic an export control directive, prompting the company to disable its most advanced artificial intelligence models for all customers. Anthropic’s dispute with the White House remains unresolved, and some of the company’s employees are still prohibited from using its most advanced artificial intelligence models.
The Trump administration’s demand that OpenAI and Anthropic limit the availability of their most advanced AI models creates an uncertain environment for other U.S. AI labs. Over the past two years, the Trump administration has sought to remove regulations and red tape that could hinder U.S. AI innovation and potentially harm the country’s competitiveness with China. However, in recent months, the White House has become increasingly concerned about the cybersecurity capabilities of recent artificial intelligence models and has made efforts to address the issue.
OpenAI plans to expand the range of customers it will be able to provide GPT-5.6 to next week, including some international partners. OpenAI executives said they couldn’t share details about how the White House approves these clients – the company simply sends the U.S. government a list and then receives feedback on it, executives said.
The White House did not immediately respond to WIRED’s request for comment.
OpenAI says its GPT-5.6 AI models will be available in three versions: Sol, the most competent version of this model; Terra, a mid-range version of the model; and Luna, a quick and inexpensive version. The company claims that GPT-5.6 Sol is its highest-performing model in its benchmarks to date testing cybersecurity, biological and agentic capabilities. In addition to these recent capabilities, OpenAI says it has a “layered security stack” aimed at stopping bad actors from exploiting its AI model for cyberattacks and other malicious behavior.