MIT AI Risk Repository

The MIT AI Risk Repository is a flagship FutureTech-linked project that aggregates and systematizes the fragmented literature on risks from artificial intelligence. Drawing on more than 74 existing AI risk frameworks and classifications, the project team has extracted and categorized over 1,700 distinct risks into an openly accessible database. Each risk entry is linked to its source documents, including paper titles, authors, and supporting evidence such as quotes and page references, enabling users to trace claims back to the original literature.

The Repository’s Causal Taxonomy classifies risks by factors such as whether they stem from human or AI actions, whether they are intentional or unintentional, and whether they occur pre‑deployment or post‑deployment. Its Domain Taxonomy groups risks into seven high-level domains and 24 subdomains (for example, misinformation and related information hazards). Together, these taxonomies provide a structured map of the AI risk landscape that can be searched, filtered, and downloaded.

The Repository is intended as a regularly updated, extensible resource that offers an accessible overview of AI threats, a common frame of reference for research and governance, and practical inputs for curricula, audits, and policy. It is positioned as part of a broader MIT AI Risk Initiative that aims to increase awareness and adoption of sound AI risk management practices across the AI ecosystem.