Stanford Institute for Human-Centered Artificial Intelligence
AI-generated text. This page was generated using artificial intelligence.
Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI) is a Stanford University institute that brings multiple disciplines into AI research, education and policy. Its work includes governance of foundation models, access to public research resources and the social and economic consequences of AI. It connects university research with policymakers through publications, training and convening.[1]
Institutional role and related centers
HAI's March 2024 institutional retrospective identifies Fei-Fei Li, John Etchemendy, Chris Manning and James Landay among its founders. It describes a university-wide approach involving technical fields, social sciences and humanities. The Digital Economy Lab and Center for Research on Foundation Models (CRFM) are distinct centers described in that account; their researchers and particular projects should receive appropriate attribution rather than having all their work treated as undifferentiated institutional advocacy.[2]
The retrospective also describes an industry affiliate program through which company employees engage with faculty, workshops and educational activities. That relationship is relevant context for HAI's university-industry work, but it does not establish that an affiliate funded or controlled a specific policy report. The source is an institutional account of these relationships, rather than an independent audit of research influence.[2]
Policymaker education and public resources
HAI's policy program includes a congressional staff boot camp, training for government employees and an online course for public servants. It also describes work on a National AI Research Resource intended to widen access to computing and government datasets. Its foundation-model governance work involves collaboration with CRFM and Stanford's Regulation, Evaluation, and Governance Lab (RegLab).[1]
These activities address different routes into policymaking: educating officials, producing empirical evidence and proposing institutional resources. A training program is not a legislative endorsement by its participants, and a university publication is not a government instruction. HAI's policy page also identifies health-care governance and analysis of executive action as continuing areas of work.[1]
Open foundation models
The December 2023 brief Considerations for Governing Open Foundation Models, co-authored by Marietje Schaake, argues that models with widely available weights can support competition, innovation and scrutiny. It recommends evaluating the additional risks of open release against closed models and technologies already available, rather than asking whether open models pose any conceivable risk.[3]
The brief warns that licensing and liability proposals may disproportionately burden open-model developers and suggests that some interventions are better directed at downstream uses. It also recognizes the difficulty of restricting misuse once model weights are released. Its account of then-pending legislative negotiations is historical; it should not be used as a substitute for the final EU AI Act. The contribution is a framework for weighing release choices and regulatory effects, not a finding that all open models are safe.[3]
Foundation Model Transparency Index
In December 2025, a research team from Stanford, Berkeley, Princeton and MIT published the third annual Foundation Model Transparency Index. The accompanying HAI article reports an average score of 40 out of 100 across 13 developers and a decline from the preceding edition. It identifies persistent gaps in disclosure about training data, computing, model use and societal effects.[4]
The index examines public information and disclosure practices, including environmental information. Its scores are transparency measurements under the researchers' methodology, not rankings of legal compliance or proof that a model is safe or unsafe. The authors argue that repeatedly opaque areas can identify priorities for policy intervention. They also distinguish release of model weights from broader transparency: open availability alone does not disclose every relevant aspect of development and deployment.[4]
Selected contributions
- Open-foundation-model governance brief (December 2023).[3]
- Policymaker education and National AI Research Resource work.[1]
- 2025 Foundation Model Transparency Index analysis (December 2025).[4]
Related articles
References
- ↑ 1.0 1.1 1.2 1.3 1.4 Stanford HAI, Policy. Accessed September 12, 2026.
- ↑ 2.0 2.1 2.2 2.3 Stanford HAI, Stanford HAI at Five, March 15, 2024. Accessed September 12, 2026.
- ↑ 3.0 3.1 3.2 Stanford HAI, Considerations for Governing Open Foundation Models, December 13, 2023. Accessed September 12, 2026.
- ↑ 4.0 4.1 4.2 Rishi Bommasani, Kevin Klyman, Alexander Wan and Percy Liang, Transparency in AI Is on the Decline, December 9, 2025. Accessed September 12, 2026.