← Back to all briefings
AI 5 min read Published Updated Credibility 91/100

NIST Launches U.S. AI Safety Institute — November 17, 2023

The National Institute of Standards and Technology established the U.S. AI Safety Institute to develop test methods, evaluations, and guidelines that implement the President’s AI Executive Order.

Kodi C.

Editor & Research Lead

Editorially reviewed for factual accuracy

AI pillar illustration for Zeph Tech briefings
AI deployment, assurance, and governance briefings

NIST launched the U.S. AI Safety Institute on 17 November 2023 as the primary federal entity for AI safety research, testing, and standards development. The institute responds to Executive Order 14110 mandates and positions the United States in international AI safety coordination alongside the UK AI Safety Institute and emerging bodies in other jurisdictions.

Institutional Mission and Structure

The AI Safety Institute operates within NIST's Information Technology Laboratory with a mandate spanning research, evaluation, and guidance development. Core functions include developing measurement science for AI systems, creating evaluation methodologies for frontier AI models, and establishing safety benchmarks for federal procurement and voluntary adoption.

The institute builds on NIST's AI Risk Management Framework (AI RMF 1.0) published in January 2023. Where the framework provides high-level governance principles, AISI focuses on technical operationalization through specific test methods, reference setups, and quantitative metrics. This progression from principles to practice mirrors NIST's approach in cybersecurity with the Cybersecurity Framework and subsequent setup guidance.

Research and Testing Priorities

Initial research priorities focus on evaluating advanced AI systems, particularly large language models and foundation models. Red-teaming methodologies receive significant attention given Executive Order requirements for developers of powerful AI systems to conduct and share results of safety evaluations. AISI will develop standardized red-teaming approaches that balance rigor with practical applicability.

Evaluation infrastructure includes reference datasets, benchmark tasks, and measurement protocols for assessing AI system behavior. Unlike commercial benchmarks focused on capability, AISI evaluations emphasize safety properties including reliability, robustness, and alignment. Measurement science addresses the fundamental challenge of quantifying concepts like trustworthiness that resist simple metrics.

Generative AI receives particular attention given rapid deployment and emerging risks. Research areas include hallucination measurement, output provenance verification, and content authenticity. Watermarking and content authentication standards support both technical development and policy setup around synthetic media.

Consortium and Collaboration Model

The AI Safety Institute Consortium brings together industry, academic, and civil society partners in pre-competitive research. Consortium membership provides access to AISI research, influence over priorities, and early visibility into emerging standards. Major technology companies, AI laboratories, and research institutions comprise initial membership.

Consortium activities include collaborative research on evaluation methodologies, shared testing infrastructure, and joint guidance development. This model enables resource pooling and knowledge sharing while preserving competitive dynamics in commercial AI development. Participation terms balance openness with intellectual property protections.

International coordination connects AISI with counterpart organizations including the UK AI Safety Institute, EU AI Office, and emerging bodies in Canada, Japan, and other jurisdictions. Bilateral and multilateral arrangements enable evaluation methodology harmonization, research collaboration, and coordinated standards development.

Enterprise Implications

Organizations developing or deploying AI systems should monitor AISI outputs for several reasons. Federal procurement now references NIST standards, and AISI evaluations may inform future requirements. Voluntary adoption of AISI methodologies shows due diligence for AI governance purposes. Early engagement with consortium activities provides influence over standards affecting commercial deployments.

AI governance programs should incorporate AISI guidance into risk assessment processes. Evaluation methodologies provide structured approaches for internal testing. Safety benchmarks offer reference points for procurement specifications and vendor assessments. Standards participation enables organizational input into requirements that may become mandatory.

Implementation Recommendations

  • Research engagement: Monitor AISI publications and explore consortium participation for organizations with significant AI investments.
  • Evaluation adoption: Incorporate AISI evaluation methodologies into internal AI testing and validation processes.
  • Standards tracking: Follow AISI standards development activities relevant to organizational AI use cases.
  • International alignment: Consider AISI guidance alongside UK and EU frameworks for multinational AI governance.
  • Procurement integration: Reference AISI benchmarks in AI procurement specifications and vendor assessments.

You may also find useful

Hand-picked articles on adjacent topics.

Continue in the AI pillar

Return to the hub for curated research and deep-dive guides.

Visit pillar hub

Latest guides

  • AI Governance Implementation Guide

    Operationalise the EU AI Act, ISO/IEC 42001, and U.S. OMB M-24-10 requirements with accountable inventories, controls, and reporting workflows.

  • AI Incident Response and Resilience Guide

    Coordinate AI-specific detection, escalation, and regulatory reporting that satisfy EU AI Act serious incident rules, OMB M-24-10 Section 7, and CIRCIA preparation.

  • AI Procurement Governance Guide

    Structure AI procurement pipelines with risk-tier screening, contract controls, supplier monitoring, and EU-U.S.-UK compliance evidence.

Coverage intelligence

Published
Coverage pillar
AI
Source credibility
91/100 — high confidence
Topics
NIST · AI Safety · Testing and Evaluation
Sources cited
3 sources (nist.gov, hitehouse.gov, iso.org)
Reading time
5 min

Documentation

  1. NIST Launches U.S. Artificial Intelligence Safety Institute — National Institute of Standards and Technology
  2. Fact Sheet: Biden-⁠Harris Administration Takes Action to Advance the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence — The White House
  3. ISO/IEC 42001:2023 — Artificial Intelligence Management System — International Organization for Standardization
  • NIST
  • AI Safety
  • Testing and Evaluation
Back to curated briefings

Comments

Community

We publish only high-quality, respectful contributions. Every submission is reviewed for clarity, sourcing, and safety before it appears here.

    Share your perspective

    Submissions showing "Awaiting moderation" are in review. Spam, low-effort posts, or unverifiable claims will be rejected. We verify submissions with the email you provide, and we never publish or sell that address.

    Verification

    Complete the CAPTCHA to submit.