AI Risk Repository Report updated (April 2025)
A consolidated, peer-informed AI risk taxonomy gives APS risk and governance teams a structured reference for developing or benchmarking agency-level AI risk frameworks.
Key points
- MIT's AI Risk Repository now covers 65 frameworks and 1,612 unique coded risk entries, updated April 2025.
- A new multi-agent risk subdomain has been added, reflecting emerging governance gaps in agentic AI systems.
- The repository supports auditing framework development and policy transparency - directly applicable to APS risk work.
Summary
MIT's AI Risk Repository has released an updated preprint (April 2025), expanding its coverage to 65 AI risk frameworks and 1,612 systematically coded risk entries. The update introduces a new subdomain on multi-agent risks, reflecting growing interest in agentic AI governance. The repository uses two taxonomies - causal and domain-based - to organise risks across academic, industry, and policy literature. It is designed as a living resource to support risk identification, auditing framework development, and policy transparency, making it a practical reference for government agencies developing or stress-testing their own AI risk approaches.
Implications for Australian agencies
- Consider APS teams developing or reviewing AI risk frameworks could assess the MIT repository's domain taxonomy and 1,612 coded entries as a benchmarking or gap-analysis tool.
- Monitor Agencies implementing or planning agentic AI use cases may want to monitor the new multi-agent risk subdomain as guidance in this area is still emerging globally.
Implications are AI-generated. Starting points, not advice.
"AI Risk Repository Report updated (April 2025)" Source: MIT AI Risk Repository – Blog Published: 23 April 2025 URL: https://airisk.mit.edu/blog/new-version-of-the-ai-risk-repository-preprint-now-available MIT's AI Risk Repository has released an updated preprint (April 2025), expanding its coverage to 65 AI risk frameworks and 1,612 systematically coded risk entries. The update introduces a new subdomain on multi-agent risks, reflecting growing interest in agentic AI governance. The repository uses two taxonomies - causal and domain-based - to organise risks across academic, industry, and policy literature. It is designed as a living resource to support risk identification, auditing framework development, and policy transparency, making it a practical reference for government agencies developing or stress-testing their own AI risk approaches. Implications for Australian agencies: - [Consider] APS teams developing or reviewing AI risk frameworks could assess the MIT repository's domain taxonomy and 1,612 coded entries as a benchmarking or gap-analysis tool. - [Monitor] Agencies implementing or planning agentic AI use cases may want to monitor the new multi-agent risk subdomain as guidance in this area is still emerging globally. Retrieved from SIMS, 18 May 2026.