CourtGuard: A Model-Agnostic Framework for Zero-Shot Policy Adaptation in LLM Safety

Umid Suleymanov, Rufiz Bayramov, Suad Gafarli +4

Current safety mechanisms for Large Language Models (LLMs) rely heavily on static, fine-tuned classifiers that suffer from adaptation rigidity, the inability to enforce new governa…