Robust LLM safeguarding via refusal feature adversarial training

Yu, Lei, Do, Virginie, Hambardzumyan, Karen · arXiv · 2024