The 30-Day Gate: White House Nears Voluntary Frontier Model Review Framework with OpenAI, Anthropic, Google

The White House is finalizing a voluntary framework requiring OpenAI, Anthropic, and Google to submit frontier models for a 30-day classified NSA review before release; announcement expected before August 1.

The framework stems from Executive Order 14409 (June 2026). Under it, developers submit cutting-edge models to federal agencies 30 days before public release; the NSA runs classified benchmarks to identify national-security risks; and a voluntary AI cybersecurity clearinghouse coordinates remediation. The review applies to frontier closed-API systems -- OpenAI, Anthropic, and Google are at the table. Meta's open-weight Llama models create a harder problem (weights can't be recalled once released), and no agreement with Meta has been reached.

Why it matters: Every major Claude, GPT, and Gemini release will face a ~30-day government preview window once the framework is signed. That's a predictable delay, not a blocker -- but if you're planning an API migration around a specific model launch date, build in buffer time. The classified benchmarking process also means security findings may never be public.

Sources: Voluntary on Paper, Mandatory in Practice -- TechTimes · Executive Order creates voluntary AI review regime -- Crowell & Moring · White House voluntary AI framework 8 key requirements