The White House is reportedly asking OpenAI to hold back the public release of its next frontier model, GPT 5.6, and instead let a small group of partners test it first. According to a report cited by TechCrunch, the move stems from safety and security concerns raised by the Trump administration.
OpenAI had not been expected to follow the usual pattern of a broad launch. Instead, the company is said to be planning a limited preview in which access is approved customer by customer before any wider release. CEO Sam Altman reportedly told employees this week that, if the preview goes well, OpenAI could open the model to a larger audience a few weeks later.
The report suggests the administration is not only reviewing the model, but working closely with OpenAI on how it will be distributed. Two government offices were named as part of the process, the Office of the National Cyber Director and the Office of Science and Technology Policy.
The White House's reported intervention reflects growing concern in Washington about powerful AI systems and their potential misuse. Although the Trump administration initially described its AI stance as hands-off, it has recently moved toward more direct involvement in model testing and oversight.
Earlier this month, President Trump signed an executive order encouraging some AI companies to voluntarily submit new systems for government review before releasing them publicly. The order marked a narrower approach than some industry critics had feared, but it still signaled a more active federal role than the administration had previously emphasized.
The reported pressure on OpenAI also appears to mirror practices already used by some AI developers. Anthropic has limited access to certain frontier models by releasing them only to selected partners. That company has argued that highly capable systems can pose risks if they are widely available too soon.
Much of the concern around advanced AI models centers on cybersecurity. Generative AI tools can already help write malware and support malicious activity, and researchers have warned that future systems may be able to identify software vulnerabilities more quickly than human analysts.
That kind of capability could create serious risks for organizations that rely on complex software systems. Hidden bugs and unpatched flaws can provide entry points into enterprise networks, making any tool that can automatically find and exploit them especially sensitive.
At the same time, because the latest models remain closed to the public, it is difficult to measure how capable they are in practice or how easily they could be abused.
For OpenAI, the reported government request could affect how one of its most advanced models reaches the market. For the administration, it would represent another step toward a more formal role in AI oversight. The reported approach also adds to the broader debate over whether limited access is a prudent safeguard or an unnecessary barrier to releasing new technology.
OpenAI has not publicly detailed a release plan for GPT 5.6, and the White House has not formally explained the reported request. But if the preview proceeds as described, it would be another sign that frontier AI launches are becoming as much a policy issue as a product decision.