White House asks OpenAI to slow roll release of new model over safety concerns
The US administration has requested OpenAI to limit access to its newest model, GPT 5.6, initially sharing it only with select partners before a broader public release.
According to The Information, OpenAI’s release of its newest model, GPT 5.6, will differ from previous launches. The company plans to initially distribute it only to a select group of close partners rather than the public, following a request from the Trump administration. At a meeting this week, CEO Sam Altman reportedly told staff that the government would be “approving access customer by customer” during a preview period. Altman added that if the limited release goes well, OpenAI hopes to follow with a general, broader release a “couple of weeks later.”
The Trump administration, which originally positioned itself as taking a “hands-off” approach to AI, has in recent months pushed for federal oversight of new models. The agencies that reportedly asked for a limited release were the Office of the National Cyber Director and the Office of Science and Technology Policy.
Earlier this month, Trump signed an executive order directing certain AI companies to voluntarily submit new models to the government for testing and evaluation before releasing them publicly.
This situation mirrors Anthropic’s controversial decision earlier this year to release its new frontier cyber model, Claude Mythos, only to a small coterie of partners through a program called Project Glasswing. Anthropic argued that the model was too powerful and could, in the wrong hands, cause more harm than good. Observers have debated whether Anthropic’s rhetoric is a marketing gimmick or a legitimate attempt to prevent misuse.
The specific concern with frontier cyber tools like Mythos is their ability to identify and exploit software vulnerabilities at speeds no human analyst could match. Since many software systems contain hidden bugs that serve as entry points into enterprise networks, this poses significant problems for organizations running complex software infrastructure. However, since these models remain closed to the public, it is difficult to assess the true extent of the threat.
