Repository Update: April 2025
A consolidated, versioned AI risk taxonomy spanning 65 frameworks gives APS risk and governance practitioners a credible reference for gap analysis and framework alignment.
Key points
- MIT AI Risk Repository Version 3 now contains over 1,600 coded AI risk categories drawn from 65 published frameworks.
- Nine new frameworks added include the final International Scientific Report on the Safety of Advanced AI and EU AI Act-aligned systemic risk taxonomies.
- A new subdomain on multi-agent risks addresses failure modes like miscoordination, conflict, and collusion in interacting AI systems.
Summary
MIT's AI Risk Repository has released Version 3, expanding its structured database to over 1,600 coded risk categories across 7 domains and 24 subdomains. Nine new frameworks have been added, including the final International Scientific Report on the Safety of Advanced AI, taxonomies aligned with EU AI Act definitions of systemic risk, and a new subdomain addressing multi-agent AI risks. The repository also incorporates practical tools such as MLCommons' AILUMINATE benchmark and a taxonomy of generative AI misuse tactics drawn from 200 real-world incidents. The resource is designed to support researchers, policymakers, and practitioners in comparing and applying AI risk frameworks.
Implications for Australian agencies
- Consider Agencies developing or reviewing AI risk frameworks may want to cross-reference their risk registers against the MIT repository to identify coverage gaps.
- Monitor The new multi-agent risks subdomain is worth watching as APS agencies begin evaluating agentic AI use cases and associated governance requirements.
Implications are AI-generated. Starting points, not advice.
"Repository Update: April 2025" Source: MIT AI Risk Repository – Blog Published: 1 April 2025 URL: https://airisk.mit.edu/blog/april-2025-update-of-the-ai-risk-repository-2 MIT's AI Risk Repository has released Version 3, expanding its structured database to over 1,600 coded risk categories across 7 domains and 24 subdomains. Nine new frameworks have been added, including the final International Scientific Report on the Safety of Advanced AI, taxonomies aligned with EU AI Act definitions of systemic risk, and a new subdomain addressing multi-agent AI risks. The repository also incorporates practical tools such as MLCommons' AILUMINATE benchmark and a taxonomy of generative AI misuse tactics drawn from 200 real-world incidents. The resource is designed to support researchers, policymakers, and practitioners in comparing and applying AI risk frameworks. Implications for Australian agencies: - [Consider] Agencies developing or reviewing AI risk frameworks may want to cross-reference their risk registers against the MIT repository to identify coverage gaps. - [Monitor] The new multi-agent risks subdomain is worth watching as APS agencies begin evaluating agentic AI use cases and associated governance requirements. Retrieved from SIMS, 18 May 2026.