OpenAI has unveiled GPT-5.6, describing it as its most capable model family yet — but most people cannot use it. The company is starting with a limited preview for a small set of partners whose access is being coordinated with the US government.
What was announced
OpenAI said the GPT-5.6 family spans several models aimed at different needs, from a high-end flagship for demanding, agent-style tasks to faster, cheaper options for everyday use. In its own safety documentation, the company rated the models "High" — a notable but not its top tier — in cybersecurity and biological/chemical capability, saying they can help find software vulnerabilities but cannot carry out autonomous, end-to-end attacks on hardened targets. Those are OpenAI's characterizations and have not been independently tested.
A restricted rollout
Rather than a public launch, OpenAI said it is "starting with a limited preview for a small group of trusted partners whose participation has been shared with the government," with a wider release expected weeks later, Digital Trends reported. The step followed a request from two White House offices — the Office of the National Cyber Director and the Office of Science and Technology Policy — to hold the model back pending a national-security review, Axios reported. Chief executive Sam Altman told staff the government would be approving access "customer by customer" during the preview.
Voluntary, for now
The review draws on a Trump administration executive order that asks AI companies to share frontier models with the government for cybersecurity review before public release — on a voluntary basis. There is no federal law compelling such pre-release vetting, which makes OpenAI's compliance a choice rather than a legal requirement. Reports said similar restrictions were applied weeks earlier to models from rival Anthropic, though the details of any government order could not be independently confirmed.
OpenAI's objection and the critics
OpenAI accepted the arrangement while publicly pushing back on it as a precedent. Altman said the company had made clear to the government that "this is not our preferred long-term model" and would work toward "a more sustainable approach." Critics went further: Representative Lori Trahan said the case-by-case gatekeeping amounted to "no law, no process, no oversight," and an analyst at the pro-AI-safety group Public First called current practice "ad hoc, personalized, opaque," The Hill reported. A White House official said the administration would keep working with AI labs on shared approaches.
Why it matters
The episode is being described as the first time the US government has pressed an American company to hold back a frontier AI model before release. Supporters of a review say powerful models warrant security scrutiny; critics say doing it without a clear legal framework invites inconsistency and opacity. Either way, GPT-5.6's staggered debut is a marker of how entangled cutting-edge AI and government have become — and a test of who decides when the most capable systems reach the public.
