US government plans voluntary rules for new AI models with tech firms

The US government is in talks with AI companies to create voluntary rules for releasing new AI models, with an announcement expected next week. The rules would set performance standards and release times for advanced models, and control who can access them, to prevent misuse.

In June, President Trump signed an order requiring federal agencies to work with top AI firms to test models before release. Google has discussed its upcoming code generation model with the government and is part of industry-wide talks.

The US Commerce Department lifted export controls on Anthropic's Fable and Mythos models on Tuesday, after suspending them three weeks ago over national security risks. OpenAI also faces restrictions, delaying the full launch of GPT-5.6 last week at the government's request, only giving access to a small group of approved partners.