OpenAI is reportedly taking a different approach with the launch of its latest model, GPT 5.6, opting to limit its availability to a select group of partners rather than a wide public release. This decision is influenced by directives from the Trump administration, according to reports.
During a recent gathering, CEO Sam Altman informed employees that the government will approve access on a case-by-case basis during an initial preview phase. He mentioned that if this limited rollout proves successful, OpenAI anticipates a broader public release a few weeks afterward.
This indicates that the Trump administration is urging OpenAI to adopt a strategy similar to that of Anthropic, which has voluntarily chosen to restrict access to its most advanced AI models.
Reports suggest that the upcoming model from OpenAI is under close government scrutiny, with staff collaborating with regulatory bodies regarding its release. The request for a restricted launch comes from the Office of the National Cyber Director and the Office of Science and Technology Policy.
Initially, the Trump administration favored a hands-off strategy toward AI but has shifted in recent months to support increased federal oversight of emerging models. Trump recently enacted an executive order that encourages specific AI companies to voluntarily submit their new models for government assessment prior to public release.
Earlier this year, Anthropic caused a stir by revealing that its advanced cyber model, Claude Mythos, would only be available to a limited group of partners through its initiative called Project Glasswing. Anthropic contended that the model’s power made it potentially dangerous in the wrong hands. This move has sparked discussions about whether the company’s statements serve as a marketing strategy or reflect a genuine effort to prevent misuse of a powerful tool.
As automated tools have long been exploited by cybercriminals, the emergence of generative AI now arms them with unprecedented capabilities. Large language models (LLMs) are particularly skilled in crafting malware, with some able to launch entire ransomware attacks autonomously.
The specific worries surrounding advanced cyber tools like Mythos stem from their potential ability to rapidly identify and exploit software vulnerabilities, far faster than a human can. Given that many software systems harbor concealed bugs that could serve as gateways into corporate networks, this represents a significant risk for organizations managing complex digital infrastructures. Nonetheless, the actual threat posed by these models remains uncertain since they are not publicly accessible.
