At White House Request, OpenAI Delays Public Release of Next-Gen AI Models

On June 26, OpenAI announced that, at the request of the Trump administration, it would delay the public release of its latest frontier AI model series—GPT-5.6, which includes Sol, Terra, and Luna. OpenAI stated that while these models have potential value for helping cybersecurity defenders, security considerations have led the company to offer only limited previews to "a small group of trusted partners," with participation details shared with the government. The company emphasized that it still plans to broadly release these models in the coming weeks, noting, "We believe in broad access, but as part of our ongoing engagement with the U.S. government, we pre-disclosed the plan and model capabilities before release."

This move follows an artificial intelligence executive order signed by Trump earlier this month, which asks some AI companies to voluntarily submit new models for government testing and evaluation before public release. Previously, Anthropic took a similar approach when releasing its frontier network model, Claude Mythos, offering access only to a select group of partners through the "Project Glasswing" initiative. Industry analysts suggest that the White House's intervention with OpenAI reflects the U.S. government's growing focus on cybersecurity risks posed by frontier AI models, particularly in the context of potential malicious use in cyberattacks.

Outlook & Analysis: OpenAI's "limited preview" strategy provides a window for government review while attempting to balance its own "open AI" mission. As the U.S. Congress advances AI safety legislation—such as requiring NIST to establish a vulnerability database for AI systems—deep collaboration between AI companies and the government before new model releases may become the norm. Finding a sustainable balance between innovation speed and safety regulation will remain a long-term challenge for the entire industry.