OpenAI has decided to roll out its newest AI model in phases following a directive from the U.S. government, a strategy reminiscent of how Anthropic introduced its Mythos product. This approach has been met with OpenAI's discontent, as they believe it restricts access to vital AI tools for “users, developers, enterprises, cyber defenders, and global partners who need them."
On Friday, OpenAI announced the debut of its GPT 5.6 series, which will initially be available to a select group of trusted partners. In a blog post, the company explained, “As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch. At their request, we are starting with a limited preview for a small group of trusted partners, whose participation has been coordinated with the government, before we broaden the release.”
Anthropic, a close competitor, had previously undergone a similar process with its Mythos model but ultimately withdrew it from public access due to government orders restricting foreign nationals from using its powerful cyber capabilities. Initially choosing to delay the launch voluntarily, OpenAI appears to have complied with the federal request.
Expressing concern over this governmental approach, OpenAI asserted, “We don’t believe this kind of government access process should become the long-term default.” Nonetheless, they view this constrained release as a necessary interim measure that will pave the way for wider accessibility in the coming weeks, as they collaborate with the White House to formulate a vetting and deployment strategy for new models—an effort that aligns with an executive order from former President Trump.
In another similarity to Anthropic’s situation, OpenAI confirmed that all entities granted access to GPT 5.6 will be based in the U.S., with plans to incorporate international partners shortly afterward. Employees of those companies from “supported countries,” such as the UK and Australia, will also be granted access.
The GPT 5.6 series includes three versions: Sol, the most powerful option; Terra, offering slightly reduced performance at a lower cost; and Luna, the most affordable variant. OpenAI claims that GPT 5.6 Sol is its “strongest model yet” while assuring it does not surpass a “cyber critical threshold” in its internal evaluation of AI capabilities. Moreover, the company emphasized that Sol is better suited for identifying and addressing vulnerabilities than for executing end-to-end cyber attacks.
In a message to employees, OpenAI CEO Sam Altman noted that the federal government requested a phased release. He explained that approvals would be granted on a customer-by-customer basis during this preview phase and suggested a broader release could happen in a few weeks if all goes smoothly.
OpenAI has made it clear to the government that it doesn’t favor this arrangement as a long-term solution, pledging to collaborate with federal agencies and industry stakeholders to establish a more sustainable model for future releases. The company has been engaged with the government regarding this model, which strengthens the underlying technology for its ChatGPT chatbot. The staggered approach emerged from discussions with the Office of the National Cyber Director and the Office of Science and Technology Policy.
Reports indicate that Trump’s commerce secretary, Howard Lutnick, sought to block even a limited rollout, insisting on approvals from other government bodies. Recently, the president signed an executive order aimed at creating a voluntary framework for vetting advanced AI models prior to their release, signaling a departure from the administration's earlier stance on deregulation in the AI sector. Just last year, Vice President JD Vance warned that excessive regulation could stifle this transformative industry. The evolving perspective reflects the increasing capabilities of models like Mythos, which have been recognized by the UK's AI security body as significant advancements over previous models.
