The US AI Safety Institute (AISI) is a state-backed organization within the National Institute of Standards and Technology (NIST) tasked with evaluating and ensuring the safety of advanced artificial intelligence (AI) models, also known as frontier AI. It was established in November 2023 during the AI Safety Summit in the United Kingdom, alongside the UK's AI Safety Institute. The institute focuses on developing measurement science, standards, and guidelines for AI safety, including pre-deployment testing of frontier models. In 2025, the US AISI was renamed the Center for AI Standards and Innovation (CAISI), reflecting its expanded role in standards-setting and innovation.
Establishment and Early Activities
The US AISI was created as part of NIST, a agency of the US Department of Commerce. Its founding followed growing concerns about AI safety, particularly after public declarations in 2023 about potential existential risks from advanced AI. During the AI Safety Summit in November 2023, the US and UK both announced the creation of their respective institutes. The US AISI's initial mission included developing evaluation methods for frontier AI models, which are often developed by companies like OpenAI, Anthropic, and Google DeepMind. In April 2024, the US and UK AISIs agreed to collaborate on at least one joint safety test, aiming to harmonize evaluation procedures. However, as reported by Politico in April 2024, many AI companies had not yet shared pre-deployment access to their most advanced models for evaluation, with some waiting for the institutes to agree on common rules.
International Network and Collaboration
In May 2024, at the AI Seoul Summit, international leaders agreed to form a network of AI Safety Institutes, including those from the UK, US, Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada, and the European Union. The US AISI participates in this network, which aims to coordinate safety evaluations and share best practices. In July 2025, the network held an exercise to explore issues with evaluating AI agents, particularly regarding leaking sensitive information or cybersecurity. Network members also met at NeurIPS 2025 in San Diego. The US AISI also collaborates with other national institutes, such as the UK's AI Safety Institute (later renamed AI Security Institute) and Japan's AISI, to advance global AI safety standards.
Role and Functions
The US AISI operates under NIST, which is known for its measurement science and standards development. The institute's key functions include conducting evaluations of frontier AI models, developing testing methodologies, and creating guidelines for AI safety and trustworthiness. It works with both government and private sector stakeholders, including AI developers and researchers. The institute also contributes to international standards efforts, aligning with NIST's broader mission. As of 2025, the institute has been renamed CAISI, but its core focus on AI standards and innovation remains. The institute is part of the broader US government strategy to manage AI risks while promoting innovation in fields like machine learning and generative AI.
Relationship with Other AI Safety Efforts
The US AISI is distinct from other national AI safety bodies, such as the UK's AI Safety Institute, which was established in London and later opened an office in San Francisco in May 2024. The US institute also coordinates with the EU AI Office, founded in May 2024, and other members of the international network. While the US AISI focuses on domestic standards and evaluations, it participates in global discussions on AI safety, including those at the AI Seoul Summit and subsequent meetings. The institute's work is complementary to that of other organizations, such as the Berkeley AI Research lab and MIT CSAIL, which conduct academic research on AI safety. However, the US AISI's primary role is regulatory and standards-based, rather than purely research-oriented.
Future Directions
As of 2025, the US AISI (now CAISI) is expected to continue its work on AI standards and evaluation, particularly as frontier AI models become more capable. The institute will likely play a key role in shaping US AI policy, especially in areas like large language models and deep learning. The renaming to CAISI suggests a broader mandate that includes fostering innovation, not just safety. The institute will also continue to participate in the international network, which may expand to include more countries. As AI technology evolves, the US AISI will need to adapt its evaluation methods to address new risks, such as those from AI agents and autonomous systems. The institute's success will depend on its ability to collaborate with industry, academia, and international partners.