OpenAI is holding back its next major model, GPT-5.6, after the federal government asked it to slow down.
CEO Sam Altman told employees during a company Q&A on Wednesday that GPT-5.6 would launch in limited preview rather than a broad release. Access will be restricted to a small set of enterprise customers, and during that preview period the Trump administration will approve individual customers before they get in. The arrangement was reported by The Information.
The more striking detail is the comparison: the source notes this deal is actually more favorable than what the Trump administration gave rival Anthropic, suggesting OpenAI's Washington relationships are paying off in practical ways — or that GPT-5.6 raised specific red flags that Anthropic's models did not. Either way, the federal government is now an active gatekeeper in how frontier AI models reach the market.
A White House with case-by-case approval authority over enterprise AI customers is a new kind of soft regulation — one that bypasses Congress entirely and leaves the rules wherever the administration decides to draw them this week.