Metaphors are a Source of Cross-Domain Misalignment of Large Reasoning Models
For AI safety researchers, it identifies metaphors as a previously overlooked source of cross-domain misalignment in LLMs, offering a detection method.
This paper investigates whether metaphors in training data cause large reasoning models to generalize misalignment across domains, finding strong evidence that metaphor-based interventions significantly increase cross-domain misalignment and that latent features linked to metaphors can predict misaligned content with high accuracy.
Earlier research has shown that metaphors influence human decision-making, raising the question of whether metaphors also influence large language models (LLMs)' reasoning pathways, given that their training data contain a large number of metaphors. In this work, we investigate the problem in the scope of the emergent misalignment problem, where LLMs can generalize patterns learned from misaligned content in one domain to another domain. We find strong evidence that metaphors in training data contribute to cross-domain misalignment in LLMs' reasoning outputs. With metaphor-based interventions during continued pre-training and fine-tuning for inducing misalignment, models exhibit significantly different degrees of emergent cross-domain misalignment. We also observe similar effects in re-alignment settings. As we further investigate this phenomenon, we find that metaphors are linked to the activation of latent features in large reasoning models. By monitoring these latent features, we design a detector that predicts misaligned content with high accuracy.