US government asks OpenAI to limit release of new GPT model

The U.S. government is intervening in the release of advanced artificial intelligence models, with a recent request to OpenAI to limit the public debut of its upcoming GPT-5.6 model.

Sources indicate that the Trump administration has asked OpenAI to provide its latest AI model only to a trusted group of partners before a wider release. This move comes just two weeks after the administration blocked foreign access to Anthropic's top-tier AI models. The U.S. government is increasingly treating advanced AI technology as a strategic national security asset, similar to semiconductors and military technology, thereby strengthening its oversight and control over these developments.

According to reports, OpenAI CEO Sam Altman has informed employees about the government's request, stating that while it's not their preferred approach, they are open to collaborating with the government on a sustainable long-term strategy. The information suggests that GPT-5.6 might initially be available to about 20 partners through Amazon Web Services' AI development platform, Bedrock.

“

We have made it clear that this is not our preferred approach, and we will work with the government and others to develop a sustainable approach going forward.

This intervention signifies an expansion of U.S. government control over AI, moving beyond semiconductors and infrastructure to encompass access to frontier models. Previously, on June 12, the administration issued export control guidelines restricting foreign access, including foreign employees of Anthropic, to its advanced models 'Mythos 5' and 'Fable 5' citing national security concerns. This action followed Amazon, a major Anthropic investor, raising security issues after Mythos 5's release.

A White House official explained to Axios that the government wants to ensure all relevant departments have tested and approved the model, emphasizing that the stricter stance is due to GPT-5.6 possessing capabilities similar to Mythos, not a sudden policy shift. However, Anthropic disputes the severity of the security issues identified by the U.S. government in Mythos 5, describing them as minor vulnerabilities. Axios noted this as the first instance of the U.S. government requesting AI companies to preemptively limit model releases before their public launch.

“

The relevant departments within the U.S. government all wanted to test and approve this model. It's not that the administration suddenly started taking a tougher stance, but rather because GPT-5.6 has capabilities similar to Mythos.