White House to finalise voluntary AI-release standards this week

The administration is in advanced talks with OpenAI, Google and Anthropic to finalise voluntary standards for testing the most capable AI models before they're released. Reporting on 2 July said an announcement could come as soon as this week. The Center for AI Standards and Innovation (CAISI) and the NSA are expected to help set the rules and monitor compliance.

'Voluntary' is backed by the implicit threat of export controls — the same lever that gave Anthropic roughly 90 minutes to pull Claude Fable 5 and Mythos 5 on 12 June, before controls lifted on 30 June. The unresolved points are the review timeline and the capability threshold that defines a 'frontier' model. Open-source advocates warn a heavy-handed reach into US labs could hand ground to China's cheaper open-weight models.

Watch for the final text and the two numbers that decide how much bite it has: the review window and the frontier threshold. It follows the 2 June executive order that set up a voluntary government-testing process while pre-empting state AI laws. This remains a developing story.

The stakes go beyond the three companies at the table. Critics warn that leaning hard on US labs does nothing to gate cheaper Chinese open-weight models, which ship freely under permissive licences and sit outside any US review — so an over-strict regime could slow American releases while doing little for actual safety. Supporters counter that a repeatable pre-launch check beats June's ad-hoc, 90-minute intervention. The two numbers still being negotiated — the review timeline and the 'frontier' threshold — will decide which camp is right.