Executive Order 14110
intermediateUS Executive Order on Safe, Secure, and Trustworthy AI (October 2023). Establishes safety testing requirements, red-teaming mandates, and reporting obligations for developers of powerful AI systems.
Overview
Executive Order 14110 is the most significant US government action on AI safety to date. It requires developers of large AI models to share safety test results with the government and establishes standards for AI security and reliability. Key provisions include: mandatory reporting for training runs above compute thresholds, red-teaming requirements before deployment, standards for authenticating AI-generated content, and guidelines for AI use in critical infrastructure. The order directs federal agencies to develop sector-specific AI guidelines and establishes NIST as a key body for AI safety standards. While not legislation, it signals serious government attention to AI risks.
Key Concepts
Compute Thresholds
Training runs above specified FLOP thresholds trigger reporting requirements.
Safety Testing Mandates
Required evaluation of dual-use capabilities before deployment.
Content Authentication
Standards for identifying AI-generated content.
Sector-Specific Guidance
Tailored requirements for healthcare, finance, infrastructure, etc.