U.S. weighs voluntary AI release standards with major tech firms

The White House is in advanced talks with major AI developers on voluntary standards for releasing new models, with a framework possible as soon as next week. The proposal would set benchmarks for model quality, release timing and who can access systems inside the United States and abroad, while Washington tries to tighten oversight of frontier AI without imposing a licensing regime.

The talks sit on top of Executive Order 14409, which Donald Trump signed on June 2 and titled Promoting Advanced Artificial Intelligence Innovation and Security. The order directs federal agencies to work with developers on a voluntary framework for covered frontier models, creates an AI cybersecurity clearinghouse and a classified benchmarking process for advanced AI cyber capabilities, and says the government should give trusted partners secure early access to strengthen cybersecurity and promote secure innovation. It also says nothing in the order should be read as authorizing mandatory licensing, pre-clearance or permitting for AI model development, publication, release or distribution.

In June, the administration was set to ask leading developers to voluntarily submit their most capable models for government cybersecurity testing up to 30 days before public release, a move Google executive Kent Walker called “an important step forward.” Google has also been in talks with the government ahead of the release of advanced coding models. Anthropic and OpenAI, both of which are preparing for initial public offerings, are part of the broader discussions as well.

On June 12, the U.S. government ordered a suspension of access to Anthropic’s Fable 5 and Mythos 5 models for any foreign national, including foreign national employees. The Commerce Department lifted those export controls on June 30 after a review, and Anthropic will restore access and redeploy Fable 5 with new classifiers and additional cybersecurity safeguards to better block misuse. OpenAI also delayed the full public launch of GPT-5.6 at the government’s request, limiting access to vetted partners.