Executive Order 14110: the first major US executive order on safe AI
In one sentence Biden signs Executive Order 14110 on AI: the largest models (10^26-operation threshold) must be reported to the government with red-team test results shared, NIST red-teaming guidelines, standards for labeling synthetic content, and measures to attract AI talent.
Until October 2023, companies developing AI in the US could release any model without demonstrating safety to any authority. Freedom was almost total.
With Executive Order 14110, President Biden changed this. It is not an act of Congress — executive orders can be faster and more direct — but it carries immediate force for all federal agencies and suppliers working with the US government.
The main points: developers of the largest models (identified by compute thresholds) must notify the government and share their safety (red-team) test results. NIST must define guidelines for AI red-teaming. The Commerce Department must prepare guidance for authenticating and labeling AI-generated content (watermarking) to counter deepfakes. Measures are included to make it easier for foreign AI experts to come to the US.
The order also covers AI in critical infrastructure, chemical and biological risks, civil rights, workers, consumers, privacy, and government AI use.
Companies
US Government
Tools
—
Tags
Sources
- https://bidenwhitehouse.archives.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/
- https://www.nist.gov/artificial-intelligence/executive-order-safe-secure-and-trustworthy-artificial-intelligence
- https://www.federalregister.gov/documents/full_text/text/2025/01/28/2025-01901.txt
- https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf