Strategic Report  ·  2026-11-20

The AI Risk Mitigation Repository

Strategic ReportHigh impactGlobal
MIT FutureTech (affiliated with MIT Sloan and CSAIL), with the University of Queensland and more than 40 collaborators including IBM Research, Stanford, MIRI, and UN University, released the AI Risk Mitigation Repository — a systematic review of 47 AI governance and risk-management frameworks cataloging 2,206 proposed AI safeguards, each tagged with the risk it addresses, who acts, when in the lifecycle, and how it reduces risk. Only 55% of mitigations target AI systems directly; the rest address organizational processes, governance structures, public policy, and broader societal systems. Responsibility is distributed across developers (~41%), deploying organizations (~33%), and governance actors (~18%). Critical coverage gaps cluster on human oversight of increasingly autonomous agents, societal resilience/recovery from AI disruptions, and compute-resource stewardship — each under 2% of documented mitigations — while organizational accountability is the largest category at 13.4%. A Delphi panel of 272 experts rated competitive dynamics and power centralization among the five risks most likely to cause severe or catastrophic harm by 2030, yet only 15 mitigation records address them. The team introduced PRIMER, an eight-item reporting checklist (mitigation, risk addressed, implementer, timing, method), noting only 3.4% of records shared consistent names.
This is the most comprehensive public map of what organizations can actually do about AI risk and where coverage is thinnest — directly usable to design an AI risk register and assurance program rather than repeat generic governance advice.
Map the PRIMER reporting checklist to your AI risk taxonomy and audit your existing risk register for the under-covered categories (agent oversight, resilience, and compute governance).
The AI Risk Mitigation Repository report PDF
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →