U.S. urges Meta to submit AI models for voluntary review

The U.S. government is urging Meta to submit its advanced artificial intelligence models for a voluntary security review, leaving the company as the only major American AI developer that has not agreed to such an arrangement. The request, communicated via email in recent weeks, seeks to evaluate the capabilities and potential vulnerabilities of Meta's technology before public release.

This push is part of a broader federal effort to increase oversight of frontier AI systems, driven by concerns that powerful models could be misused for cyberattacks or military purposes. The framework for these reviews was established by an executive order signed by President Donald Trump on June 2. The order invites developers to offer their "covered frontier models" to the government for up to 30 days of evaluation before releasing them to trusted partners and gives officials until the end of July to formalize the review process.

While the process is not yet official, several of Meta's competitors have already been cooperating. OpenAI and Anthropic have been working with the government to test unreleased models. In May, Google DeepMind, Microsoft, and xAI agreed to provide the Center for AI...