OpenAI is releasing GPT-5.6 to a small set of partners, with access cleared by the US government "on a customer by customer basis," according to a staff memo reported by The Information.
The arrangement came out of talks with the Office of the National Cyber Director and the Office of Science and Technology Policy, and traces back to a Trump executive order calling for a "voluntary" cybersecurity review of new models. Sam Altman told staff the company had made clear this "is not our preferred long term model," and Commerce Secretary Howard Lutnick reportedly phoned to warn against shipping without sign-off from more agencies. The top tier, Sol, is said to have cleared 96.7% of a cyberattack benchmark, crossing the risk threshold that triggered the caution.
Voluntary review now looks a lot like de facto licensing: a federal desk deciding who may buy a commercial model, with no waitlist and no public date for wider release.