White House keeps AI oversight framework secret after briefing leading tech firms

Framework reportedly targets the most advanced AI models but excludes open-weight systems

The White House is opting not to publish a newly completed framework designed to assess the cybersecurity risks posed by the most advanced AI models.

According to the WSJ, officials shared an overview of the voluntary oversight system with representatives from OpenAI, Anthropic, Meta, Google, Nvidia and other leading AI developers during a meeting at the White House on Tuesday.

However, the administration has decided not to release the framework publicly, at least for now.

The framework was completed by an 1st August deadline set by President Donald Trump in an executive order signed in June, a White House official said.

Some industry figures fear that keeping the framework confidential could discourage participation by companies that were not invited to Tuesday's briefing because they have little information about how the review process would operate.

Closed models targeted

The proposed framework reportedly focuses on the most capable proprietary AI models developed by US companies, particularly those that demonstrate advanced cybersecurity or hacking capabilities based on government performance benchmarks.

Developers of those closed models would be invited to voluntarily submit their systems for government testing before public release.

Open-weight models, which can be downloaded, modified and run independently by users, would reportedly fall outside the framework. Those systems have become increasingly popular among researchers, start-ups and major technology companies that argue open development encourages innovation.

Nvidia's chief executive Jensen Huang has repeatedly argued that open-weight AI is essential to maintaining US competitiveness. Nvidia also develops AI software alongside the chips that power many of the world's leading AI systems.

Balancing security and innovation

The administration's approach comes amid growing debate over how governments should regulate increasingly capable AI systems without slowing technological progress.

Recent disclosures by OpenAI and Anthropic that experimental models had exhibited unexpected behaviour during testing, including attempts to compromise external systems, have heightened concerns about the cybersecurity risks posed by frontier AI.

At the same time, the rapid emergence of increasingly capable open-weight models from Chinese developers has intensified competition between Washington and Beijing.

Some officials have reportedly discussed imposing restrictions on certain Chinese AI companies over allegations that their models benefited from advances made by US competitors.

Several executives at OpenAI and Anthropic have publicly supported stronger action against Chinese AI firms.

Trump has said the US must avoid regulations that could leave it trailing China in AI.

Speaking in the Oval Office last week, he said the administration needed to strike a balance between protecting national security and ensuring the US remained competitive in the global AI race.

The question of whether open-weight AI models should face tighter controls has become one of the most contentious issues in US technology policy.

More than 80 technology companies recently signed an open letter organised by Nvidia urging the US government to support open-weight AI development.

On Tuesday, Nvidia and its partners also launched Shared AI Findings Exchange (SAFE), an initiative intended to allow companies to confidentially share information about AI-related incidents and identify recurring safety issues.

Hugging Face, Red Hat and the Linux Foundation are among the organisations backing the project.