@misc{indiciaeb8d4957ca09f,
  title = {From Specialization to Generalization: Instruction-tuned LLMs for Robust Harmful Content Mitigation},
  author = {Lukas Edman and Daryna Dementieva and Alexander Fraser},
  year = {2026},
  url = {https://arxiv.org/abs/2608.25605},
  note = {Source identifier: 2608.25605}
}