The White House has asked OpenAI to initially release its next artificial intelligence model, GPT-5.6, to a small group of government-approved partners rather than to the public at large, as US officials move to tighten scrutiny of powerful AI systems, Axios reported.
The request came from the White House’s Office of the National Cyber Director and the Office of Science and Technology Policy, according to the report, as the Trump administration works on a framework to test and evaluate the security risks posed by so-called frontier AI models.
If carried out, it would mark the first time the US government has sought to influence the rollout of a leading AI model before its public release. Axios reported that OpenAI has been coordinating with the administration on plans for GPT-5.6, though OpenAI Chief Executive Sam Altman told employees that a restricted launch was “not our preferred long-term model” and that the company would seek a more sustainable approach for future releases.
The administration’s interest stems from concerns that GPT-5.6 could have “Mythos-like” capabilities, a reference to an advanced cybersecurity-focused AI model developed by Anthropic that was not released publicly because of its potential power.
According to Axios, US officials have already been briefed on GPT-5.6’s capabilities. Commerce Secretary Howard Lutnick has discussed the model with Altman and pushed for relevant government agencies to test and approve it before a broader rollout.
The move follows an executive order signed earlier this month by President Donald Trump directing federal agencies to establish a voluntary testing protocol for frontier AI models prior to release. Internal debates over how restrictive such a framework should be delayed the order for weeks, the report said.
Anthropic’s Mythos model, unveiled in April, was designed to identify previously unknown cybersecurity vulnerabilities before they could be exploited by hackers. Anthropic opted against a public launch, instead granting limited access to select technology firms under its Project Glasswing initiative, which it said was aimed at securing critical software systems worldwide. |
