The White House is asking OpenAI to slow roll the release of its new model over safety concerns

OpenAI is set to approach the launch of its latest model, GPT 5.6, quite differently than before. Instead of a public rollout, the company plans to provide access to a select group of trusted partners. This decision comes in response to requests from the Trump administration, as reported by The Information. During a recent meeting, OpenAI's CEO, Sam Altman, informed employees that the government would be vetting access to the model on a case-by-case basis during an initial preview phase. He mentioned that if this limited launch is successful, OpenAI aims to expand access to a broader audience in a few weeks. This shift reflects the administration’s increasing insistence on regulating AI technologies. Previously characterized by a more hands-off approach, the Trump administration has recently advocated for federal oversight of new AI models. Earlier this month, an executive order was signed that requires certain AI firms to voluntarily submit their models for government evaluation before public release. The Office of the National Cyber Director and the Office of Science and Technology Policy are among the agencies that have expressed concerns over the release, indicating a desire for more controlled distribution of powerful AI technologies. Meanwhile, Anthropic has already taken similar precautions, limiting access to its advanced cyber model, Claude Mythos, through a program called Project Glasswing. The company justified this choice by citing the potential risks of powerful AI tools in the wrong hands, raising questions about whether this is a genuine effort to mitigate harm or merely a marketing strategy. The rise of generative AI has transformed how cybercriminals operate, equipping them with sophisticated tools capable of writing malware and executing ransomware attacks autonomously. With frontier cyber models like Mythos able to swiftly identify and exploit software vulnerabilities, the implications for organizations with complex software systems could be severe. However, the current limited access to these models makes it challenging to fully assess the extent of the threat they may pose, leaving many in the industry to ponder the balance between innovation and safety.