@article{whensafetyfailsbeforetheanswerbenchmarki,
  title = {When Safety Fails Before the Answer: Benchmarking Harmful Behavior Detection in Reasoning Chains},
  author = {Ishita Kakkar and Enze Zhang and Rheeya Uppaal and Junjie Hu},
  year = {2026},
  eprint = {2604.19001},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2604.19001},
}