A Comprehensive Evaluation of LLM Unlearning Robustness under Multi-Turn Interaction

Machine unlearning aims to remove the influence of specific training data from pre-trained models without retraining from scratch, and is increasingly important for large language models (LLMs) due to safety, privacy, and legal concerns. Although prior work primarily evaluates unlearning in static, single-turn settings, forgetting robustness under realistic interactive use remains underexplored. In this paper, we study whether unlearning remains stable in interactive environments by examining two common interaction patterns: self-correction and dialogue-conditioned querying. We find that knowledge appearing forgotten in static evaluation can often be recovered through interaction. Although stronger unlearning improves apparent robustness, it often results in behavioral rigidity rather than genuine knowledge erasure. Our findings suggest that static evaluation may overestimate real-world effectiveness and highlight the need for ensuring stable forgetting under interactive settings.

Key Contributions

Identifies two realistic interaction patterns (self-correction and dialogue-conditioned querying) that recover knowledge appearing forgotten under static single-turn evaluation.
Demonstrates that stronger unlearning methods produce behavioral rigidity rather than genuine knowledge erasure, and that static evaluation systematically overestimates real-world unlearning effectiveness.
Proposes interactive evaluation paradigms as a necessary complement to single-turn benchmarks for assessing LLM unlearning robustness.