The Trump administration is requesting that Meta submit its artificial intelligence models to the federal government for voluntary review, according to people familiar with the matter who spoke with news outlets. The request is intended to permit government officials to examine both the capabilities and any vulnerabilities of the models.
People familiar with the discussions said Meta - the parent company of Facebook - remains the only major U.S. developer of AI technology that has not consented to voluntarily sharing its models with the federal government for review. Meta introduced its most advanced model, Muse Spark, in April.
Officials say the voluntary review process would give government evaluators the opportunity to test unreleased or advanced systems in order to better understand what the models can do and where weaknesses might exist. The exact scope and format of the requested review were not described by those cited.
Earlier this month, Washington ordered Anthropic to suspend access to its most advanced AI models for foreign nationals, citing security concerns. Anthropic complied by removing the models entirely from public access. Separately, OpenAI and Anthropic had been working with the U.S. government to test unreleased AI systems. In addition, Google’s DeepMind, Microsoft, and xAI have recently agreed to undergo national security evaluations.
The request to Meta follows a pattern of U.S. officials seeking to coordinate with AI developers on assessments intended to surface capabilities and potential risks. The reporting cited four people with knowledge of the matter.
Summary
The administration has asked Meta to voluntarily submit its AI models for federal review so officials can evaluate their capabilities and vulnerabilities. Meta has not agreed to such sharing and launched its Muse Spark model in April. Other companies, including Anthropic, OpenAI, Google’s DeepMind, Microsoft, and xAI, have engaged with government testing or evaluations in recent weeks.
Key points
- U.S. officials have requested that Meta provide voluntary access to its AI models for government review.
- Meta is reported to be the only major U.S. AI developer that has not agreed to voluntary model sharing; Muse Spark was launched in April.
- Other developers - including Anthropic, OpenAI, DeepMind, Microsoft, and xAI - have cooperated with or been subject to government testing or restrictions.
Risks and uncertainties
- Uncertainty over whether Meta will agree to the voluntary review request and on what terms - this affects the technology sector and regulators engaged in national security oversight.
- Ongoing national security concerns that have already led to restrictions on access to certain models, as illustrated by Washington's order to Anthropic - this has implications for firms developing advanced AI models and for related markets.