OpenAI rolled out its new GPT-5.6 model on Friday, but only in a restricted form. The release comes after the White House asked the company to cap early access to a small group of vetted partners while officials evaluate the model's security profile. Three versions launched, named Sol, Terra, and Luna, with the most capable variant going first to roughly twenty approved customers, including access routed through Amazon's Bedrock platform. OpenAI said the government is approving partners on a case by case basis and expects a fuller public release within a few weeks.
This matters because it shows Washington now treats top-tier AI releases the same way it treats sensitive exports, not just software updates. The same pattern played out weeks earlier with Anthropic Fable 5 and Mythos 5, both pulled back over concerns that their cyber capabilities could be misused. OpenAI is no longer an exception to that scrutiny, it is the second case in a row.
The risk for the industry is that ad hoc government reviews, done without a finished legal framework, could become the default way frontier models reach the market, slowing competition and giving early movers in government approval a structural edge. Watch for how fast OpenAI's "few weeks" promise holds, and whether the pending cyber executive order framework formalizes this review process for every future model launch.
