Key facts
- The U.S. government is urging Meta to voluntarily submit its AI models for review.
- Meta plans to automate up to 90% of its risk assessments using AI by May 31, 2025.
- Concerns have been raised by current and former employees about increased risks and reduced scrutiny.
- Meta claims human expertise remains for novel issues, with automation limited to low-risk decisions.
The U.S. government is urging Meta Platforms to voluntarily submit its artificial intelligence models for review to evaluate their capabilities and potential vulnerabilities. This push comes as Meta plans to automate up to 90% of its risk assessment processes by replacing human reviewers with AI, aiming to streamline the release of new features and algorithm updates. However, this move has sparked concerns among current and former Meta employees who fear that relying on AI for critical decisions about privacy, safety, and content moderation could lead to unforeseen real-world harm and reduced scrutiny. Meta states that human expertise is still utilized for novel and complex issues, with automation primarily applied to low-risk decisions. Internal documents suggest the company is considering automating reviews for sensitive areas including AI safety, youth risk, and content integrity.