The US AI Safety Institute (AISI) is a state-backed organization within the National Institute of Standards and Technology (NIST) that evaluates and promotes the safety of advanced artificial intelligence (AI) models, often called frontier AI. It was created in November 2023 during the AI Safety Summit in the United Kingdom, alongside the UK's AI Safety Institute, in response to growing concerns about potential existential risks from AI. In 2025, the institute was renamed the Center for AI Standards and Innovation (CAISI), reflecting a broader focus on standards and innovation.
The institute operates as part of NIST, a non-regulatory agency of the US Department of Commerce. Its primary mission is to develop measurement science, standards, and evaluation methods for AI safety, including testing frontier models for risks such as bias, misuse, and cybersecurity vulnerabilities. It collaborates with other government agencies, academic institutions, and private companies to advance AI safety research and practice.
Establishment and early activities
The US AISI was announced on 1 November 2023, at the AI Safety Summit at Bletchley Park, where the UK and US both committed to creating national AI safety institutes. The US institute was launched under NIST, with initial funding and staff drawn from existing NIST AI programs. In April 2024, the US and UK AISIs agreed to collaborate on at least one joint safety test, aiming to harmonize evaluation procedures. This partnership was part of a broader effort to set international standards for AI safety, as many AI companies had not yet shared pre-deployment access to their most advanced models for evaluation.
International network and collaboration
At the AI Seoul Summit in May 2024, international leaders agreed to form a network of AI Safety Institutes, including the US, UK, Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada, and the European Union. The US AISI participates in this network, which conducts joint exercises and shares best practices. In July 2025, the network held an exercise to explore issues with evaluating AI agents, particularly regarding sensitive information leakage and cybersecurity. Members also met at NeurIPS 2025 in San Diego to discuss AI safety evaluation methods.
The US AISI also engages with international partners bilaterally. For example, it has worked with the UK AISI to align evaluation standards, and it has participated in discussions with the EU AI Office and other national institutes to develop common frameworks for frontier AI testing.
Focus areas and key products
The institute focuses on several key areas: developing evaluation benchmarks for large language models and other generative AI systems, creating guidance for AI risk management, and conducting research on AI safety in collaboration with academic labs like MIT CSAIL and Stanford AI Lab. It also works on standards for AI transparency, accountability, and robustness, often in coordination with NIST's broader AI initiatives.
Key products include technical reports on AI model evaluations, public datasets for safety testing, and frameworks for assessing AI risks. The institute also contributes to international standards efforts, such as those led by the International Organization for Standardization (ISO), to ensure that AI safety practices are globally consistent.
Relationship with industry and academia
The US AISI collaborates with leading AI companies, including OpenAI, Anthropic, and Google DeepMind, to gain access to frontier models for pre-deployment testing. It also partners with academic institutions to advance research on AI safety, such as Berkeley AI Research and Carnegie Mellon University. These partnerships aim to bridge the gap between theoretical research and practical safety evaluations.
In 2024, the institute faced challenges in obtaining pre-deployment access to some advanced models, as reported by Politico. However, it continued to develop voluntary evaluation programs, encouraging companies to submit models for testing. The institute's work is complemented by other government efforts, such as the White House's AI Bill of Rights and executive orders on AI safety.
Evolution to CAISI
In 2025, the US AISI was renamed the Center for AI Standards and Innovation (CAISI), signaling a shift toward a broader mandate that includes promoting innovation alongside safety. This change aligns with the institute's growing role in setting standards for AI evaluation and its participation in international networks. The renaming reflects an ongoing evolution in how governments approach AI governance, balancing safety with economic competitiveness.
As of 2025, CAISI continues to operate under NIST, with a focus on developing standards, tools, and evaluations to ensure that AI technologies are safe, secure, and trustworthy. Its work remains critical as AI systems become more capable and integrated into society.