White House reportedly pushes OpenAI to limit GPT 5.6 release over safety concerns
White House asks OpenAI to slow-roll GPT 5.6
The White House is reportedly pressing OpenAI to delay the public release of its next model, GPT 5.6, over safety concerns, according to The Information. Instead of a broad launch, OpenAI is said to be preparing a limited release to a small group of close partners while the model is reviewed by the Trump administration.
At a meeting this week, OpenAI CEO Sam Altman reportedly told staff that access would be “approving customer by customer” during a preview period. If the rollout goes smoothly, Altman reportedly said the company hopes to follow with a wider release a couple of weeks later.
The reported request comes from the Office of the National Cyber Director and the Office of Science and Technology Policy, both of which are said to have worked closely with OpenAI on the upcoming model. The arrangement would mark a notable shift from OpenAI’s previous public launches, and it mirrors the kind of restricted access Anthropic has already used for some of its most powerful systems.
The Trump administration, which initially cast itself as taking a hands-off approach to artificial intelligence, has recently moved toward more direct oversight. Earlier this month, Trump signed an executive order directing certain AI companies to voluntarily submit new models for government testing and evaluation before public release.
The reported caution reflects growing concern around frontier AI systems with cyber capabilities. Generative AI tools have already been used to write malware, and some can carry out parts of ransomware attacks autonomously. The worry with newer frontier cyber models is that they may be able to identify and exploit software vulnerabilities at a much higher level than earlier systems.
Anthropic drew attention earlier this year when it limited access to its frontier cyber model, Claude Mythos, through a partner-only program called Project Glasswing. The company said the model was too powerful to release broadly. That position drew skepticism from some observers, though the broader debate over how to balance capability and misuse has only intensified as more advanced models approach release.
Sources: