The US AI Safety Institute (AISI) is a state-backed organization established in November 2023 to evaluate and ensure the safety of advanced artificial intelligence (AI) models, also known as frontier AI models. It operates under the National Institute of Standards and Technology (NIST), a federal agency within the US Department of Commerce. The institute was created during a period of heightened global attention to AI safety, following public declarations in 2023 about potential existential risks from AI. In 2025, the US AISI was renamed the Center for AI Standards and Innovation (CAISI), reflecting a shift toward standards-setting and innovation support. The institute is part of an international network of AI safety institutes, which includes counterparts in the UK, Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada, and the European Union, as agreed at the AI Seoul Summit in May 2024.
The primary mission of the US AISI is to conduct independent evaluations of advanced AI models, develop safety guidelines, and promote best practices among AI developers. This includes assessing risks such as bias, misuse, and catastrophic harms, and providing technical guidance to federal agencies and industry. The institute collaborates with academic institutions, industry partners, and international counterparts to advance AI safety research and policy. Its work is critical in an era where Artificial intelligence systems, including Large language models and Generative AI models, are rapidly deployed across sectors.
Establishment and Early Activities
The US AISI was officially established during the AI Safety Summit held in November 2023, alongside the UK's AI Safety Institute. The summit, hosted by the UK Prime Minister Rishi Sunak, aimed to position the UK as a global leader in AI safety regulation. Sunak emphasized the need for independent safety evaluations, stating that AI companies cannot "mark their own homework." The US AISI was created as part of NIST, leveraging the agency's expertise in measurement and standards. In April 2024, the US and UK concluded an agreement to collaborate on at least one joint safety test, marking a significant step in international cooperation. However, a Politico report from the same month noted that many AI companies had not shared pre-deployment access to their most advanced models for evaluation, highlighting challenges in enforcement.
Role and Responsibilities
The US AISI focuses on evaluating frontier AI models, which are the most capable and potentially risky systems. Its responsibilities include developing evaluation methodologies, conducting safety tests, and issuing guidance for model developers. The institute also works on mitigating risks associated with Machine learning and Deep learning technologies, including issues like model robustness, transparency, and alignment with human values. It collaborates with other federal agencies, such as the White House Office of Science and Technology Policy, to coordinate AI safety efforts across the government. The institute's work is informed by research from leading AI labs, including OpenAI, Anthropic, and Google DeepMind, as well as academic institutions like MIT CSAIL and Stanford AI Lab.
International Collaboration
As part of the international network of AI safety institutes, the US AISI engages in joint exercises and knowledge sharing. In July 2025, the network held an exercise to explore issues with evaluating AI agents, particularly regarding leaking sensitive information or cybersecurity threats. Members also met at NeurIPS 2025 in San Diego, a major conference for Neural network research. The US AISI has bilateral agreements with several countries, including the UK, Japan, and South Korea, to harmonize evaluation standards and share technical expertise. This collaboration is essential for addressing global AI risks, as AI models are developed and deployed across borders.
Evolution to CAISI
In 2025, the US AISI was renamed the Center for AI Standards and Innovation (CAISI). This change reflects a broader mandate to not only ensure safety but also promote innovation and the development of AI standards. CAISI continues to operate under NIST, with a focus on creating voluntary guidelines and frameworks that can be adopted by industry. The renaming aligns with the US government's approach to AI regulation, which emphasizes flexibility and market-driven solutions over strict mandates. CAISI works closely with industry stakeholders, including major tech companies like Amazon Web Services, Microsoft Azure, and Google Cloud, to ensure that safety measures are practical and do not stifle technological progress.
Impact and Challenges
The US AISI has had a significant impact on the AI safety landscape, influencing both domestic and international policies. Its evaluations have informed decisions about model deployment and have prompted companies to improve their safety practices. However, the institute faces challenges, including limited resources and the rapid pace of AI development. Critics argue that voluntary guidelines may not be sufficient to prevent harm, while others worry about overregulation. The institute's effectiveness depends on its ability to adapt to new technologies, such as Transformer (architecture) architectures and Multi-Head Attention mechanisms, which underpin modern AI systems. As of 2025, the institute continues to evolve, with ongoing efforts to strengthen its evaluation capabilities and expand its international reach.
Related Developments
Other countries have established similar institutes, creating a global network. For example, the UK AISI, initially based in London, opened an office in San Francisco in May 2024 to be closer to AI companies. Japan's AISI, founded in February 2024, is part of the Information Technology Promotion Agency and employs about 23 people. South Korea's AISI, established in November 2024, is based in Bundang District and is supported by an investment of 10 to 20 million South Korean won per year. These institutes share best practices and collaborate on joint projects, such as the July 2025 exercise on AI agents. The US AISI's role in this network is central, given the country's leadership in AI development and its extensive resources.
Future Directions
Looking ahead, the US AISI (now CAISI) is expected to focus on emerging challenges, such as the safety of AI agents and autonomous systems. It will likely develop new evaluation methods for models that can take actions in the real world, such as those used in Waymo self-driving cars or Tesla. The institute may also address issues related to AI in critical sectors like healthcare and finance, where failures could have severe consequences. As AI technology advances, the need for robust safety measures will only grow, making the institute's mission increasingly important. The international network of AI safety institutes will continue to play a key role in coordinating global efforts, ensuring that the benefits of AI are realized while minimizing risks.
Conclusion
The US AI Safety Institute, now the Center for AI Standards and Innovation, represents a critical component of the US government's strategy to manage AI risks. Established in 2023, it has evolved to meet the challenges of a rapidly changing technological landscape. Through its evaluations, guidelines, and international collaborations, the institute aims to ensure that advanced AI systems are developed and deployed safely. While challenges remain, the institute's work is essential for building trust in AI and protecting society from potential harms. As AI continues to permeate every aspect of life, the mission of the US AISI will remain a cornerstone of responsible AI governance.