OpenAI's newest flagship artificial intelligence model, GPT-5.6 Sol, is being released under an unusual restriction: access during its initial preview is limited to a small group of trusted partners approved by the Trump administration. According to the article, OpenAI previewed the model's capabilities with the US government ahead of launch and, at the government's request, is beginning with a narrow rollout whose participants have been shared with officials before wider availability. This marks a notable shift, in that the question of who is granted access first — and how those decisions are made — has become as significant as when the general public can use the tool.

GPT-5.6 Sol is part of OpenAI's new GPT-5.6 line-up, which also includes models named Terra and Luna, and is described by the company as its strongest yet, with improvements in coding, biology and cybersecurity tasks. The cybersecurity dimension is what has drawn Washington's attention: OpenAI says Sol is better at helping to find and fix software vulnerabilities than at reliably carrying out full cyberattacks, and maintains that it does not cross the company's internal "Cyber Critical" threshold, while conceding that benchmarks cannot predict every use when the model is combined with other tools. Initially the GPT-5.6 models will be offered via the API and Codex to selected partners, with broader availability planned for ChatGPT, Codex and the API; as of early July the model was not yet available in ChatGPT and no general-availability date had been announced. OpenAI chief executive Sam Altman has voiced concerns about the government selecting customers for AI tools, citing national security.