OpenAI Releases New AI Models, but Not for Public Use
▼ Summary
– OpenAI is delaying the public release of GPT-5.6 at the request of the Trump White House, first sharing it with a small set of government-approved customers.
– OpenAI considers the delay temporary, hoping for broader availability in weeks, but dislikes the process, stating it should not become the long-term default.
– The delay follows a Trump executive order addressing cybersecurity concerns, creating a voluntary model-sharing process that OpenAI says does not yet exist.
– The White House also recently directed Anthropic to limit its AI model access, creating an uncertain environment for US AI labs amid growing cybersecurity concerns.
– GPT-5.6 will come in three versions (Sol, Terra, Luna), with Sol being OpenAI’s most capable model on cybersecurity, biology, and agentic benchmarks.
OpenAI has confirmed it will hold back the public launch of its next-generation AI models, GPT-5.6, following a request from the Trump administration. The company announced Friday that it will initially offer the technology only to a select group of pre-approved customers, vetted by the U.S. government. From there, OpenAI plans to gradually expand access in coordination with the White House.
The decision has frustrated internal teams, according to a source familiar with the company’s position, though OpenAI views the delay and government approval process as a temporary hurdle. In a blog post, the company expressed hope that GPT-5.6 will be available to all users within weeks. The Information first reported on the delayed rollout.
“We don’t believe this kind of government access process should become the long-term default,” OpenAI wrote. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.”
Earlier this month, President Trump signed an executive order targeting the cybersecurity risks posed by powerful new AI systems. The order outlined a “voluntary process” for AI labs to share their models with the government 30 days before a wide release. It included a carveout to prevent the process from becoming a de facto licensing system. However, during a Friday briefing, OpenAI executives noted that no such voluntary framework has been established yet. That leaves frontier AI labs in a strange interim state, where cooperation with the government on model launches appears far from optional.
The White House’s request to OpenAI comes just two weeks after it directed Anthropic to restrict access to its most advanced AI models. That directive forced Anthropic to take those models offline for all customers. The dispute remains unresolved, and some Anthropic employees are still barred from using the company’s most capable systems.
This push from the Trump administration to limit availability of cutting-edge AI models creates uncertainty for other U. S. labs. Over the past two years, the administration has worked to cut regulation that could slow American AI innovation and hurt competitiveness with China. But in recent months, the White House has grown increasingly alarmed about the cybersecurity risks tied to new AI models and has scrambled to respond.
OpenAI plans to expand the group of customers approved to use GPT-5.6 next week, including some international partners. Company executives said they cannot disclose how the White House is approving these customers. OpenAI simply submits a list of proposed users to the government and receives feedback.
The White House did not respond to WIRED’s request for comment.
OpenAI will release GPT-5.6 in three versions: Sol, the most capable; Terra, a mid-tier option; and Luna, a fast and affordable variant. The company says GPT-5.6 Sol is its most powerful model to date on benchmarks for cybersecurity, biology, and agentic capabilities. Alongside these new features, OpenAI has implemented a “layered safeguard stack” designed to prevent malicious use, including cyberattacks.
(Source: Wired)
