White House AI Framework Nears Aug. 1 Deadline

The executive order gave federal agencies 60 days to develop the implementation framework, but CNBC reported that the order itself contained limited operational detail. The White House directed CNBC to a statement from spokesperson Kush Desai saying, "Trump White House to meet a deadline we set for ourselves."

Politico reported that the draft framework would ask companies to voluntarily submit advanced models to the government 30 days before release for safety testing, followed by limited initial rollouts through government-approved partners before wider availability. The report said OpenAI, Anthropic, and Google submitted proposed edits to a draft framework, citing three anonymous sources familiar with the discussions.

Altman told CNBC that he had reviewed a draft and did not have "anything specifically in mind" regarding proposed changes. He also told CNBC that meetings in Washington covered OpenAI's upcoming models and the United States' position in the global AI race.

One central question is which systems fall under the framework. Axios reported that the administration was weighing a proposal supported by OpenAI and Anthropic to cover highly capable models under the same rules regardless of whether they are open- or closed-source. Axios attributed that account to a source familiar with the discussions.

That distinction can affect how evaluations are performed, reproduced, and monitored. Public-policy discussions around comparable systems commonly separate pre-deployment testing from post-release monitoring, particularly where model weights or APIs may enable substantially different degrees of access.

Axios also reported that European and UK experience points to scenario-specific evaluations, such as testing a model's contribution to biological-threat or government-database attack capabilities. Chris Canal, co-founder of AI evaluation firm EquiStamp, told Axios that researchers can use historical threat data to assess whether AI materially expands a bad actor's capabilities. The outlet also cited independent verification and standardized benchmarks as elements emphasized in European approaches.

For AI teams building or deploying frontier systems, the framework's final language could determine what evidence is expected before release: model documentation, capability evaluations, red-team results, access controls, or external review. Those requirements have not been publicly detailed in the reporting reviewed here.

Across comparable regulatory regimes, implementation details often determine the practical burden more than high-level commitments. Definitions of an "advanced" or "highly capable" model, treatment of open weights, evaluator independence, and handling of sensitive findings are likely to be consequential for developers, cloud providers, and organizations that deploy frontier-model APIs.