What happened
The Trump administration confirmed it finalized (by the August 1, 2026 deadline set in Executive Order 14409, 'Promoting Advanced Artificial Intelligence Innovation and Security,' signed June 2, 2026) a voluntary framework — developed by NSA, CISA, Treasury and the AI Standards and Innovation Center — for classifying 'covered frontier models' and assessing their cybersecurity/offensive-cyber capabilities before release. On August 3, 2026, a White House official confirmed a staff-level meeting with representatives of OpenAI, Anthropic and Google was scheduled for August 4 to review the framework. The content of the framework, including the classified benchmarking criteria, has not been publicly released.
Why it matters
This is the first US government-wide structured process for evaluating frontier AI models' cyber-offensive capabilities pre-release, offering developers up to a 30-day voluntary pre-release government review window. Although voluntary and largely classified, it establishes a template that could evolve into a de facto licensing/preclearance gate for frontier model releases in the US, directly affecting how OpenAI, Anthropic, Google and other frontier labs plan model launches.
Action needed
Frontier AI developers should monitor for publication of framework criteria and evaluate whether to participate in the voluntary pre-release review window; track outcomes of the August 4, 2026 White House meeting for scope clarifications.