The White House is set to announce new AI standards as soon as next week, following months of closed-door negotiations with Anthropic, OpenAI, and Google, according to the Financial Times. The framework has already been rehearsed through recent enforcement actions, including a Commerce Department order on June 12 that cut off access to Anthropic's Claude Fable 5 and Claude Mythos 5 for all foreign nationals, and OpenAI's delayed launch of GPT-5.6 at the U.S. government's request.
Under the proposed rules, frontier AI models would face a 30-day "Model Customs" review before being allowed into a trusted deployment lane. The framework stems from a June executive order that creates a voluntary "covered frontier model" designation, granting government access for up to 30 days prior to public release — effectively a pre-release security vetting period for the most advanced systems.
This structure mirrors existing export controls and controlled rollouts, suggesting a system already tested before being formalized. The government's ability to restrict access, as seen with Anthropic's 18-day model shutdown, points to enforcement mechanisms operating ahead of any official policy announcement. Three labs are understood to be involved in the initial framework discussions.
The practical impact would bind leading AI developers to a regulatory checkpoint before deploying models that could pose national security risks. While labeled "voluntary," the precedent of government-requested delays and access cuts raises questions about how voluntary the system truly is. Companies may face pressure to comply or risk harsher unilateral action.
Critics argue the framework could slow U.S. AI innovation without congressional authorization, and the line between voluntary and mandatory remains blurry. Some researchers warn the 30-day review may be too short for thorough safety assessments, while others fear it is long enough to give government overreach a toehold in emerging technology markets.