Taxonomy of Risks Posed by Language Models
A structured LLM risk taxonomy offers APS teams a ready-made reference for risk categorisation when developing AI governance frameworks or use-case assessments.
Key points
- MIT AI Risk Repository spotlights a 2022 Google DeepMind taxonomy of LLM risks across six domains and 20 subdomains.
- The taxonomy covers discrimination, information hazards, misinformation, malicious use, HCI harms, and socioeconomic harms - directly relevant to APS AI risk assessment work.
- The underlying paper is from 2022; the MIT blog post is a summary spotlight, not new research.
Implications for Australian agencies
- Consider APS teams developing AI risk registers or use-case assessment frameworks could consider referencing Weidinger et al.'s taxonomy as a structured baseline for LLM-specific risk categorisation.
- Monitor Practitioners tracking the MIT AI Risk Repository may want to monitor the broader repository for other included frameworks that complement or update this 2022 taxonomy.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.
View original source
Copied.
"Taxonomy of Risks Posed by Language Models"
Source: MIT AI Risk Repository – Blog
Published: 12 May 2025
URL: https://airisk.mit.edu/blog/taxonomy-of-risks-posed-by-language-models
The MIT AI Risk Repository's blog has spotlighted a 2022 ACM FAccT paper by Weidinger et al. (Google DeepMind) that provides a comprehensive taxonomy of ethical and social risks associated with large language models. The taxonomy spans six domains - discrimination, information hazards, misinformation, malicious uses, human-computer interaction harms, and environmental and socioeconomic harms - with 20 subdomains. It distinguishes between observed and anticipated risks, and focuses on raw LLMs rather than specific applications. The spotlight is a summary of an existing paper rather than new primary research.
Implications for Australian agencies:
- [Consider] APS teams developing AI risk registers or use-case assessment frameworks could consider referencing Weidinger et al.'s taxonomy as a structured baseline for LLM-specific risk categorisation.
- [Monitor] Practitioners tracking the MIT AI Risk Repository may want to monitor the broader repository for other included frameworks that complement or update this 2022 taxonomy.
Retrieved from SIMS, 18 July 2026.