Meta Faces White House Push to Join AI Pre-Release Review Pact

When Trump signed an executive order on June 2, 2026 establishing a voluntary 30-day pre-release review window for frontier AI models, the arrangement came with a notable gap. Meta, among the largest AI developers in the world, had not agreed to participate. According to The New York Times, the White House is now pressing Meta directly to submit its models for government testing before public release.

The evaluations run through the Commerce Department's Center for AI Standards and Innovation -- CAISI -- which operates the TRAINS program, or Testing Risks of AI for National Security. The program assesses frontier models for risks including cybersecurity, biosecurity, and chemical weapons threats. OpenAI and Anthropic agreed to similar terms in 2024 under the Biden administration. Google DeepMind, Microsoft, and xAI followed in May 2026, extending the participating lab count to five -- with Meta the notable absence.

The voluntary structure of the framework is exactly where this becomes a policy question. AI companies are not required to submit models for government testing; those that participate have done so by agreement. That makes Meta's holdout consequential: the review structure now spans its major peers but not Meta, and reportedly that gap has drawn direct White House attention.

What the reporting does not give you is why Meta has declined or what any eventual agreement would specifically cover. A voluntary framework that a major developer can opt out of without consequence raises a real question about how durable this approach actually is, and whether the administration's preference for pressing rather than mandating is sufficient to close the gap.

If Meta does eventually sign, it would complete the set of major U.S. frontier AI labs under the same pre-release evaluation structure, and push the harder policy conversation toward how voluntary review can extend across genuinely different models of AI distribution.