Trump admin asks OpenAI to limit GPT-5.6 rollout

The White House asked OpenAI to limit the initial release of GPT-5.6 to a small set of government-approved partners while federal agencies test the model under a new voluntary framework. The request came from the Office of the National Cyber Director and the Office of Science and Technology Policy.

People familiar with the discussions described the request as prompted by GPT-5.6’s capabilities, which they compared to those in recent high-end models. Officials requested restricted access while they develop procedures to evaluate advanced systems before wider deployment.

The action follows an earlier federal order that halted public access to two models from another company over national security concerns. An executive order issued earlier this month directed agencies to establish a voluntary testing program for advanced AI systems; agencies are still deciding how to structure the program and which models should undergo evaluation.

Industry leaders have pressed for stronger oversight of frontier models. In 2023, OpenAI chief executive Sam Altman told lawmakers that ‘independent oversight would eventually be necessary.’ Anthropic’s chief executive has argued that the most capable models should undergo government-backed evaluation because of risks such as enabling sophisticated cyberattacks or aiding dangerous biological research.

In a June letter to congressional staff, Anthropic alleged that operators tied to an affiliate of Alibaba generated more than 28.8 million exchanges with its Claude chatbot in an effort to extract capabilities. The company urged lawmakers to strengthen protections against large-scale model distillation.

Some experts warn that giving the largest developers a leading role in writing rules could produce uneven enforcement that favors established firms and raises barriers for smaller competitors. Regulators and lawmakers are weighing how to combine safety testing, transparency and independent review while preserving competition.

Federal officials are continuing talks with OpenAI and other companies as the testing framework is finalized. It remains unclear how long the limited rollout will last or when a broader release might be approved; any wider distribution will depend on the completion of agency evaluations and further White House guidance.