Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment
A structured LLM alignment taxonomy offers a reference point for APS agencies developing AI risk assessment frameworks or evaluation criteria.
Key points
- A 2023 academic paper proposes a taxonomy of 7 major LLM trustworthiness categories covering 29 subcategories.
- The MIT AI Risk Repository spotlights this as one of 30 risk frameworks it has catalogued - useful for APS risk inventory work.
- The paper itself is two years old; the blog post adds no new analysis beyond the repository spotlight.
Implications for Australian agencies
- Consider Agencies developing AI risk registers or evaluation criteria could consider whether this taxonomy's seven dimensions map usefully onto their existing risk categorisation structures.
- Monitor Policy teams tracking the MIT AI Risk Repository may want to monitor the full set of 30 frameworks it has catalogued for emerging patterns in AI risk classification.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.
View original source
Copied.
Appeared in:
Weekly digest, 16 February 2026
"Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment"
Source: MIT AI Risk Repository – Blog
Published: 22 February 2026
URL: https://airisk.mit.edu/blog/trustworthy-llms
The MIT AI Risk Repository has spotlighted a 2023 academic paper by Liu et al. proposing a comprehensive taxonomy for evaluating LLM trustworthiness across seven dimensions: reliability, safety, fairness, resistance to misuse, explainability and reasoning, social norms, and robustness. Each dimension contains subcategories covering risks such as hallucination, sycophancy, prompt attacks, and cultural insensitivity. The paper also provides a guideline for multi-objective evaluation using automated and templated question generation. The blog post itself is a brief summary adding no original analysis; the signal value is primarily the taxonomy structure as an input to AI risk or evaluation frameworks.
Implications for Australian agencies:
- [Consider] Agencies developing AI risk registers or evaluation criteria could consider whether this taxonomy's seven dimensions map usefully onto their existing risk categorisation structures.
- [Monitor] Policy teams tracking the MIT AI Risk Repository may want to monitor the full set of 30 frameworks it has catalogued for emerging patterns in AI risk classification.
Retrieved from SIMS, 18 July 2026.