mlsanityData to 7 Oct 2026

Metaphors are a Source of Cross-Domain Misalignment of Large Reasoning Models

arXiv 2601.03388 ↗First posted 6 Jan 2026cs.CL cs.AI cs.LGcited by 0

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 16 of its 37 references that are in the map. First posted on arXiv 6 Jan 2026.

Builds on

16 of 37 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025Qwen3 Technical Report14 May

Other references · 15

  1. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  2. 2022Emergent Abilities of Large Language Modelscited by 2,483
  3. 2024LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Modelscited by 1,442
  4. 2022Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasetscited by 599
  5. 2025Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMscited by 158
  6. 2025Natural Emergent Misalignment from Reward Hacking in Production RLcited by 111
  7. 2025Model Organisms for Emergent Misalignmentcited by 84
  8. 2025Persona Features Control Emergent Misalignmentcited by 76
  9. 2025Thought Crime: Backdoors and Emergent Misalignment in Reasoning Modelscited by 57
  10. 2024Erasing Conceptual Knowledge from Language Modelscited by 34
  11. 2025Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLMscited by 17
  12. 2025In-Training Defenses against Emergent Misalignment in Language Modelscited by 14
3 more
  1. 2025Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Modelscited by 14
  2. 2025From Narrow Unlearning to Emergent Misalignment: Causes, Consequences, and Containment in LLMscited by 5
  3. 2025RePO: Understanding Preference Learning Through ReLU-Based Optimizationcited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Science is Exploration: Computational Frontiers for Conceptual Metaphor Theory
  2. 2025What makes Reasoning Models Different? Follow the Reasoning Leader for Efficient Decoding
  3. 2025Probing Semantic Alignment, Lexical Invariance, and Syntactic Influence in LLM Metaphor Processing
  4. 2024A Dual-Perspective Metaphor Detection Framework Using Large Language Models
  5. 2022MiQA: A Benchmark for Inference on Metaphorical Questions
  6. 2025Modeling Understanding of Story-Based Analogies Using Large Language Models
  7. 2024A Nurse is Blue and Elephant is Rugby: Cross Domain Alignment in Large Language Models Reveal Human-like Patterns
  8. 2022Psychologically-informed chain-of-thought prompts for metaphor understanding in large language models