Trump AI Testing Framework Excludes Open Models, Leaves 30-Day Review Terms Undefined
Updated
Updated · The Verge · Aug 5
Trump AI Testing Framework Excludes Open Models, Leaves 30-Day Review Terms Undefined
3 articles · Updated · The Verge · Aug 5
Summary
Open AI models are entirely outside the Trump administration’s voluntary cybersecurity testing framework, which also says it cannot be used to restrict those models after release.
The finalized plan gives the government 30 days to review only closed-source frontier models, yet reportedly never defines core triggers such as “state-of-the-art” or “national security risk.”
Anthropic, OpenAI and Google were briefed at the White House on the framework, but the administration does not plan to publish the details publicly.
The framework follows Trump’s June executive order seeking pre-release access to advanced models, but its voluntary and ambiguous design could leave smaller AI providers guessing how to stay in compliance.