The U.S. government told OpenAI to hold back GPT-5.6 until federal agencies say it is ready to ship.
During a staff Q&A, CEO Sam Altman confirmed that GPT-5.6 is live only for a small group of customers handpicked by the federal government. The Office of the National Cyber Director and the Office of Science and Technology Policy asked OpenAI to stagger the rollout, and Commerce Secretary Howard Lutnick called Altman personally to warn against releasing it without prior government approval. Altman said in a memo he hoped the general release would follow within a couple of weeks. OpenAI went along but noted this "is not our preferred long-term model."
The episode mirrors what happened to Anthropic earlier this year. Anthropic staged its own careful rollout of Fable 5, a frontier model built with added safeguards, only to have the government place it on an export control list three days after launch. Foreign nationals, including Anthropic's own employees, were banned from access; Anthropic could not enforce compliance and pulled the model from the market entirely. If that is the ceiling Anthropic hit, OpenAI is staring at the same one.
A Trump executive order signed this month now requires AI labs to hand the government 30-day advance access to frontier models before any public release, making delays like this structural rather than exceptional. Neil Chilson, former FTC chief technologist and head of AI policy at the Abundance Institute, called the escalating intervention "horrible for the broader AI ecosystem," warning that arbitrary export controls will push labs to slow-walk releases and keep powerful tools out of public hands.
An administration that promised to deregulate AI has quietly built its own gatekeeping layer instead.