Can Mechanistic Interpretability Help Explain LLM Agent Failures? A Case Study on Web Element Hallucination
arXiv cs.AI · 14 Nov 2026, 04:30 AEST
Mechanistic Interpretability (MI) has made significant progress in explaining the internal computations of Large Language Models (LLMs). However, these advances have yet to demonstrate comparable success in explaining the decision-making processes of LLM-based agents.
Key facts
- This paper examines the challenges MI tools face when applied to realistic agentic failures.
- We focus on one such failure: web agents hallucinating actions that are not available on a website.
- We find that a linear probe trained on the last prompt token can predict hallucinations before generation begins (0.80 AUC) and steering along a probe direction corrects 85% of hallucinations, but disrupts 7% of valid actions.
- The probe direction is sensitive to prompt formatting, losing much of its steering effect after a minor change to how actions are represented.
- We discuss how these observations challenge the assumptions underlying standard MI methods for circuit discovery and feature identification.
Read aloud
Mechanistic Interpretability MI has made significant progress in explaining the internal computations of Large Language Models L L M s. However, these advances have yet to demonstrate comparable success in explaining the decision-making processes of L L M-based agents. This paper examines the challenges MI tools face when applied to realistic agentic failures. We focus on one such failure: web agents hallucinating actions that are not available on a website. We find that a linear probe trained on the last prompt token can predict hallucinations before generation begins zero point eight zero AUC and steering along a probe direction corrects eighty five percent of hallucinations, but disrupts seven percent of valid actions. The probe direction is sensitive to prompt formatting, losing much of its steering effect after a minor change to how actions are represented. We discuss how these observations challenge the assumptions underlying standard MI methods for circuit discovery and feature identification.
Sources in this cluster
- Can Mechanistic Interpretability Help Explain LLM Agent Failures? A Case Study on Web Element Hallucination
arXiv cs.AI · 14 Nov 2026, 04:30 AEST