In a staff meeting this week, OpenAI's CEO Sam Altman delivered news that would have been almost unthinkable a year ago: the US government would be "approving access customer by customer" before his company's most powerful new model reaches the public. That model, GPT-5.6, will first be shared with a select group of partners — not because OpenAI chose that path independently, but because the Trump administration reportedly directed it to. The decision represents a striking reversal for an administration that originally promised a "hands-off" approach to artificial intelligence regulation.

Earlier this month, President Trump signed an executive order asking certain AI companies to submit new models for government testing and evaluation before any public release. The agencies now involved in vetting GPT-5.6 include the Office of the National Cyber Director and the Office of Science and Technology Policy, both of which reportedly worked closely with OpenAI staff during the model's development. If the limited release proceeds without incident, Altman told employees, a broader rollout could follow within "a couple of weeks.

" The move aligns OpenAI, perhaps reluctantly, with a posture that its rival Anthropic had already adopted voluntarily. Earlier this year, Anthropic announced that its frontier cyber model, Claude Mythos, would be restricted to a narrow circle of partners through a programme called Project Glasswing. The justification was sobering: the company argued that a sufficiently capable model, in the wrong hands, posed dangers that outweighed its benefits. Observers remain divided over whether that framing reflects genuine caution or sophisticated marketing.