Executive Order 14110 on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence
Executive Order 14110 directs US federal agencies to govern AI safety, security, and civil rights, and requires the most powerful foundation-model developers to report safety-test results. It is policy guidance rather than statute, enforced through existing agency authorities.
Overview
Executive Order 14110 was issued by the President of the United States in October 2023. It is the most comprehensive US federal policy action on artificial intelligence to date. The order does not create new statutory law; instead it directs federal agencies to use existing authorities to manage the risks and capture the benefits of AI. It built on the earlier Blueprint for an AI Bill of Rights and the NIST AI Risk Management Framework.
The order frames AI as a technology with transformative potential and serious risks, including threats to civil rights, privacy, national security, and consumer protection. It sets a government-wide agenda spanning safety testing, critical-infrastructure protection, fraud detection, immigration, education, and federal procurement.
Who It Applies To
The order binds federal agencies directly. It reaches the private sector indirectly through two channels. First, the Defense Production Act is invoked to require developers of the most powerful dual-use foundation models to report safety-test results and red-teaming outcomes to the government. Second, agencies must shape procurement, grants, and sector regulation in line with the order, which affects any company selling AI to or operating in regulated US markets.
Key Requirements
Developers of large dual-use foundation models above defined compute thresholds must notify the government and share results of safety tests. NIST is tasked with developing standards for red-teaming and secure AI development. The Department of Commerce must develop guidance for content authentication and watermarking of AI-generated media. Agencies must appoint Chief AI Officers, inventory AI use cases, and apply safeguards when AI affects rights or safety. The order also addresses privacy-enhancing technologies, AI talent immigration, and protections against algorithmic discrimination in housing, benefits, and hiring.
Penalties for Non-Compliance
Because it is an executive order rather than a statute, it carries no direct civil or criminal penalties for private parties. Enforcement flows through existing agency authorities: failure to file required model reports can trigger Defense Production Act consequences, and agencies may condition contracts, grants, or approvals on compliance. Note that the order's status has been subject to change under subsequent administrations, and organizations should track its current force.
How to Comply
Organizations should map AI systems against the NIST AI Risk Management Framework, document model capabilities and compute usage, and maintain red-teaming and evaluation records. Foundation-model developers should establish reporting workflows for safety results. Companies selling AI to the federal government should align with emerging procurement and content-provenance guidance, and implement watermarking or provenance metadata for generative outputs where feasible.