United States - AI Development and Use (EO 14110)

Overview

Executive Order 14110, signed on October 30, 2023, set a cross-cutting federal policy to advance the safe, secure, and trustworthy development and use of artificial intelligence (AI). The Order directed an interagency program of standards development, testing, workforce expansion, and procurement reform intended to mitigate risks from high-capability and dual-use models while preserving U.S. innovation capacity. The Order emphasized development of companion resources to the NIST AI Risk Management Framework (AI RMF) for generative AI, creation of testbeds and model-evaluation tools at the Department of Energy (DOE) and the National Science Foundation (NSF), and agency-level guidance for civil rights, labor, and responsible procurement. The White House archived text of EO 14110 remains the authoritative record of the policy and assignments to agencies; that text and related agency deliverables inform the technical and governance outputs initiated by the Order. Despite the Order's later rescission by a January 2025 executive action, the EO shaped NIST and agency activities related to trustworthy AI and testing infrastructure.

Definitions

EO 14110 used and built on commonly used federal definitions for terms such as "artificial intelligence," "generative AI," "foundation models," and "dual-use models," often cross-referencing statutory and agency definitions (for example, drawing on definitions in 15 U.S.C. and agency guidance). Critical conceptual distinctions in the Order included the notion of "dual-use foundation models" (large models with generalized capabilities that may be deployed across multiple contexts and potentially weaponizable or risk-bearing), "generative AI" (models that produce novel content), and "national security systems" (which are treated separately). The Order also referenced the NIST AI RMF trustworthiness characteristics—validity, reliability, safety and security, transparency, and privacy—placing them at the center of federal technical guidance and evaluation planning.

Governance and Institutional Framework

EO 14110 assigned specific roles and timelines to a set of executive-branch offices and agencies to coordinate policy and technical work. The Office of Science and Technology Policy (OSTP) and the Office of Management and Budget (OMB) were charged with high-level coordination and prioritization; NIST was tasked with technical standard-setting and development of companion resources to the AI RMF for generative AI; the Department of Energy (DOE) and the Director of National Intelligence (DNI) were given responsibilities related to model evaluation tools and capability assessments; the National Science Foundation (NSF) was directed to accelerate research institutes and the NAIRR pilot; the Department of Justice (DOJ) and the Civil Rights enforcement community were asked to emphasize non-discrimination and equity in agency deployments; the Office of Personnel Management (OPM) and the General Services Administration (GSA) were given workforce and acquisition duties. Implementation relied on interagency working groups and consultations with industry, academia, and civil society. For full text and assignments, see the White House archived EO 14110.

Key Focus Areas

The Order’s substantive focus areas were broad and technical. Safety and risk management: directed development of testing, evaluation and red-teaming guidance, and creation of tools and testbeds (DOE, NSF) to evaluate capabilities that could produce hazardous outputs (e.g., biological, chemical, cyber, critical-infrastructure risks). Standards and measurement: asked NIST to produce companion materials to the AI RMF for generative AI and to coordinate standards development and benchmarking across agencies and with private-sector stakeholders. Transparency and disclosure: encouraged improved incident disclosure and measurement practices to support oversight and auditability. Privacy and civil rights: instructed agencies to consider impacts on civil rights (DOJ) and worker protections (Labor, OPM) and to issue guidance for agency use of AI in contexts such as criminal justice. Workforce and research capacity: tasked NSF and DOE to expand research institutes, NAIRR piloting, and training programs to grow public-sector AI talent. Procurement and access: directed GSA and OMB to facilitate agency access to commercial AI capabilities while pairing such access with governance measures. Cybersecurity and model security: directed development of secure software development companion guidance for generative and foundation models, and prioritized model-security considerations across evaluation efforts. International engagement and cooperation were highlighted insofar as standards alignment and secure supply chain work required collaboration with allies and international fora.

Implementation Framework

Implementation relied on staggered deliverables and standing interagency coordination: many sections of the Order required deliverables within 45, 120, 180, 270, or 540 days. OSTP and OMB coordinated cross-cutting action plans; NIST led technical companion resources to the AI RMF; DOE and NSF developed model-evaluation tools and testbeds and were authorized to consult private labs and third-party evaluators. Agencies were asked both to develop internal guidance for their own AI use and to support industry standards and voluntary conformity approaches. The Order emphasized use of existing statutory authorities and procurement levers (e.g., acquisition guidance via GSA) rather than creation of new standalone punitive regimes; it also encouraged independent third-party evaluation and information sharing consistent with privacy and security law. For technical baseline, agencies were instructed to leverage the NIST Resource Center and previously published AI RMF materials (NIST AI Standards and AI RMF).

Monitoring and Evaluation

EO 14110 required reporting and monitoring mechanisms: agencies were to submit plans, assessments, and periodic reports to OSTP and OMB; DOE and NSF were to develop evaluation tools capable of assessing near-term extrapolations of model capabilities; NIST and interagency groups were to develop benchmarks and auditing guidance. Monitoring emphasized measurable outcomes—benchmarks, testbeds, and audit reports—and encouraged public-private cooperation with safeguards to protect sensitive information and national-security concerns. The Order also proposed pilot programs and staged scaling to evaluate efficacy before broad deployment. Ongoing monitoring responsibilities were embedded in agency deliverables and interagency working groups to allow iterative updates to guidance and standards.

Penalties, Liability, and Appeals

EO 14110 did not itself create new criminal penalties or a novel civil enforcement regime. Instead, it relied on existing agency authorities and administrative levers—procurement preferences, grant and funding conditions, guidance, and regulation—to incentivize compliance. Liability for harms continued to be governed by existing statutory and common-law regimes (consumer protection laws, sectoral regulators, tort law), and agencies were directed to consider legal implications, including worker-protection laws and civil-rights enforcement. The Order anticipated use of administrative remedies and programmatic adjustments rather than new criminal sanctions. Where third-party evaluations or data-sharing were required, agencies were to use existing legal authorities and protections to govern appeals, redress, and classification of sensitive information.

Relationship to Other Instruments

EO 14110 explicitly connected to pre-existing instruments: it built on NIST’s AI RMF (NIST AI 100-1) and the National Artificial Intelligence Initiative, and it referenced sectoral responsibilities and statutes governing privacy, national security, and research funding. The Order aligned U.S. technical and governance work with international standardization initiatives (e.g., OECD, ISO/IEC) and the U.S. government’s broader AI strategy. It did not preempt sector-specific statutes or independent regulator actions; instead, it sought to coordinate federal activity and to equip agencies with technical tools and guidance that could feed into future rulemaking by sectoral regulators.

International Alignment

The Order emphasized international coordination for standards and measurement, urging alignment with allies and participation in multilateral fora. It called for agency engagement with international standards bodies and referenced ongoing U.S. contributions to OECD AI principles, the Global Partnership on AI, and work under the U.S.-EU Trade and Technology Council. The EO recognized that effective technical measurement and risk management require cross-border collaboration on definitions, metrics, and testing protocols to avoid fragmentation and to support interoperability in global markets.

Implementation Timeline

Sources and References