OpenAI has delayed the public release of its latest AI model, GPT-5.6, at the request of the US government, in a move that could affect how future AI tools are developed and shared.

In a bid to address cybersecurity concerns over powerful new AI models, President Trump signed an executive order earlier this month, requiring labs like OpenAI and Anthropic to share their models with the government 30 days before broader release. For now, customers must be preapproved by the US administration.

OpenAI hopes to make GPT-5.6 available to a wider audience in the coming weeks but is navigating an uncertain landscape, given unresolved issues between Anthropic and the White House. The delay highlights the tension between fostering innovation and ensuring security in the rapidly evolving AI sector.

The new models are described as having three versions: Sol, the most advanced; Terra, mid-tier; and Luna, a cost-effective option. OpenAI claims GPT-5.6 Sol is its most capable model yet on benchmark tests of cybersecurity, biology, and agentic abilities, with additional safeguards in place to prevent misuse.

This development underscores the growing role of government oversight in AI innovation, potentially setting a precedent for future releases and use cases.