US Executive Order on Safe AI (EO 14110)

Active
high risk
United States

October 2023 — Active (ongoing agency rule-making)

Official Text

high risk framework

Primarily: US frontier AI developers (>10^26 FLOPS training). Secondary: Federal agencies. Voluntary best practice for all US organizations.

Overview

Executive Order 14110 on Safe, Secure, and Trustworthy AI, signed October 2023. Requires safety testing reports for frontier AI systems, establishes federal AI governance, and directs agencies to develop sector-specific guidance.

Key Requirements

  • Safety testing reports for models trained with >10^26 FLOPS
  • Red team evaluations shared with the US government
  • NIST development of frontier AI safety guidelines
  • Federal agencies to implement AI governance policies
  • Screening for biological and cybersecurity risks
  • Watermarking standards for AI-generated content
  • Workforce impact assessments for federal AI deployment
  • Privacy-preserving AI techniques encouraged

Implementation Guidance

  1. 1Determine if training compute exceeds 10^26 FLOPS threshold
  2. 2Prepare safety test results documentation for DHS/Commerce reporting
  3. 3Implement red team evaluation processes
  4. 4Monitor NIST for new AI safety standards under the EO
  5. 5Federal contractors: review agency-specific AI policies

Penalties for Non-Compliance

Regulatory and contractual enforcement. Federal contractor debarment possible for non-compliance.

Framework Details

Short Name

US AI EO

Jurisdiction

United States

Status

Active

Risk Level

high

Enforcement Date

October 2023 — Active (ongoing agency rule-making)

Affected Organizations

Primarily: US frontier AI developers (>10^26 FLOPS training). Secondary: Federal agencies. Voluntary best practice for all US organizations.

Tags

GovernmentFrontier AIEnterpriseUSSafety Testing

This is educational guidance only. Always consult qualified legal counsel for compliance decisions affecting your organization.