MIT AI Risk Repository

The MIT AI Risk Repository is a research program housed within MIT FutureTech, an interdisciplinary group operating under MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Initiative on the Digital Economy. The project was launched in August 2024 with the release of a preprint paper on arXiv and a publicly accessible website at airisk.mit.edu. The repository set out to consolidate and clarify how risks from artificial intelligence are categorized across the fragmented landscape of academic, industry, and policy frameworks. The core database was initially built by reviewing 43 published AI risk frameworks and extracting 777 distinct risks. Each entry is linked to its source paper (with title, authors, quotes, and page numbers) and classified using two complementary taxonomies. The Causal Taxonomy organizes risks by entity (AI, human, or other), intentionality (intentional, unintentional, or unspecified), and timing (pre- or post-deployment). The Domain Taxonomy organizes risks into seven domains and 24 subdomains: discrimination and toxicity, privacy and security, misinformation, malicious actors and misuse, human-computer interaction, socioeconomic and environmental impacts, and AI system safety, failures, and limitations. The repository is described as a living resource and has been updated quarterly. Version 2 was released in December 2024 (adding 13 new frameworks), Version 3 in April 2025 (adding 9 frameworks and approximately 600 new risk categories), and Version 4 in December 2025 (adding 9 more frameworks and approximately 200 new categories, bringing the total to over 1,700 coded risks from more than 65 frameworks). The website received approximately 90,000 hits in its first year and is linked to by roughly 2,000 other websites. Several governments and large companies have incorporated it into their AI risk management processes. Related projects include the AI Incident Tracker, which uses a large language model to classify reports from the AI Incident Database according to the repository's risk taxonomy and a harm severity rating system. The team also produces governance landscape mapping and is developing the AI Risk Index, which assesses gaps between expert AI risk management recommendations and actual practices at over 200 model developers, enterprises, and governance organizations. The core team is led by Alexander Saeri (Project Director), Peter Slattery (Engagement Lead and Research Scientist), Jess Graham (Research Officer), Michael Noetel (Research Methods Specialist), Simon Mylius (Incident Tracking and Governance Mapping Lead), Neil Thompson (MIT FutureTech Director and Project Supervisor), and Robert Gambee (Software Engineer). The project also engages a large network of collaborators from MIT, the University of Queensland, Oxford, Georgetown, the Center for Security and Emerging Technology, the OECD, and other institutions. MIT FutureTech is supported by funders including Open Philanthropy, the National Science Foundation, Accenture, IBM, the MIT-Air Force AI Accelerator, and MIT Lincoln Laboratory.