protologue

Over-Refusal

Also called exaggerated safety, false refusal, overrefusal.

Over-refusal is a model declining safe requests because they superficially resemble harmful ones, such as refusing to explain how to kill a Python process.

Description

It is the cost side of safety training, since overly cautious models frustrate users and withhold legitimate help. Test suites such as XSTest pair safe prompts with unsafe look-alikes to measure the balance, and methods such as deliberative alignment aim to reduce refusals while improving robustness.

Sources

  1. Röttger et al. (2023). XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models.

Cite this entry

Protologue. (2026). Over-Refusal. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0176). https://protologue.com/t/over-refusal/

BibTeX
@misc{protologue_over_refusal,
  title = {Over-Refusal},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0176},
  url = {https://protologue.com/t/over-refusal/}
}

Markdown JSON