Cambridge, MA, Nov. 12, 2026 (GLOBE NEWSWIRE) — As artificial intelligence becomes increasingly embedded in workplaces and everyday life, organizations face an urgent question: What practical steps can they take to reduce AI-related risks? While governments, researchers, and companies have proposed thousands of safeguards for AI systems, those recommendations are spread across dozens of frameworks and reports, making them difficult to compare, evaluate, and implement.

Researchers at MIT FutureTech, which is affiliated with the MIT Sloan School of Management and the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), worked with the University of Queensland and more than 40 collaborators, including IBM Research, Stanford University, the Machine Intelligence Research Institute, and the United Nations University to address this challenge. Together, they created the AI Risk Mitigation Repository, a free public database that catalogs 2,206 proposed AI safeguards drawn from 47 existing AI governance and risk-management frameworks. 

“Organizations across sectors increasingly recognize the need to address AI risks, but it’s often unclear what action that should translate to in practice,” said Neil Thompson, principal research scientist and director of MIT FutureTech. “Without a common way to organize and compare proposed safeguards, it becomes difficult to coordinate efforts, evaluate options, or identify what’s still missing.”

The research team developed a framework for organizing each safeguard according to the risk it addresses, who is responsible for taking action, when action should occur during the AI lifecycle, and how the safeguard is intended to reduce risk. The result is a shared framework that allows organizations to more easily compare, explore, and understand approaches to AI risk management.

“People are talking a lot about AI risks, but there’s been much less clarity about what organizations can actually do about them,” said Peter Slattery, research scientist at MIT FutureTech and a contributor to the study. “We wanted to create a practical resource that helps decision-makers understand what options exist, where attention is concentrated, and where critical gaps remain.”

What did researchers discover about AI risk management?

One of the study’s central findings is that reducing AI risk extends far beyond improving the technology itself. While 55 percent of proposed mitigations focus directly on AI systems, researchers found that many others target organizational processes, governance structures, public policy, and broader societal systems.

The study also found that responsibility for managing AI risks is distributed across the AI ecosystem. Developers account for roughly 41 percent of proposed safeguards, while organizations deploying AI systems account for 33 percent. Governance actors, including regulators and policymakers, are responsible for another significant share of proposed interventions.

These findings suggest that effectively managing AI risks requires coordinated action from technology developers, businesses, governments, and civil society, rather than relying on technical solutions alone.

Where are the biggest gaps in AI safeguards?

Despite documenting more than 2,200 proposed safeguards, the researchers found several areas receiving little attention.

Among the least represented categories were proposals related to human oversight of increasingly autonomous AI agents, improving society’s resilience and recovery from AI-related disruptions, and stewardship of the computing resources that power advanced AI systems. Each accounted for less than 2 percent of documented mitigations. By comparison, organizational accountability and responsibility represented the largest category at 13.4 percent.

Researchers involved in the project identified these underrepresented areas as important opportunities for future work, particularly as AI systems become more capable and widely deployed.

The team also developed a reporting framework called PRIMER to help organizations describe AI safeguards more consistently and transparently. The framework is intended to make it easier to compare approaches across organizations and ultimately help researchers determine which mitigations are most effective in practice.

How can organizations use the AI Risk Mitigation Repository?

The AI Risk Mitigation Repository is part of the MIT AI Risk Initiative, an effort led through MIT FutureTech to improve understanding of AI risks and accelerate adoption of effective safeguards. The initiative previously launched the AI Risk Repository, which cataloged potential AI risks. This new effort focuses on identifying and organizing proposed solutions. It provides one of the most comprehensive efforts to date to organize proposed approaches for reducing AI risks. 

“Our earlier work focused on understanding AI risks,” said Slattery. “This project focuses on solutions. We found thousands of proposed ways to reduce AI risks, but we also identified important areas where relatively little attention has been paid.” 

The team identified and categorized 2,206 risk mitigations from 47 frameworks and analyzed them across multiple dimensions, including who should implement them, what type of change they seek to create, the risks they address, and the stage of the AI lifecycle where they apply. It covers a broad spectrum of interventions: some focus on technical improvements to AI systems, while others involve organizational policies, industry standards, audits, transparency measures, public oversight, regulations, and international coordination.

The researchers also developed an interactive explorer that allows users to search more than 2,200 proposed safeguards, browse them by category, and examine how they relate to different AI risks, responsible actors, and stages of the AI lifecycle. Users can also compare mitigations across different taxonomies to identify patterns, areas of concentration, and gaps in the proposed approaches to managing AI risks.

Companies can use the repository to identify safeguards relevant to their AI systems and support internal risk-management efforts. Policymakers can explore different governance and regulatory approaches across a range of AI risks, while researchers can use the data to identify understudied areas and investigate which safeguards are actually effective.

“By making these safeguards easier to find and compare, we hope organizations can make more informed decisions about managing AI responsibly,” Slattery said.

Attachment 


Matthew Aliberti
MIT Sloan School of Management
7815583436
malib@mit.edu

Media gallery

About The Author
MIT Sloan School of Management