Uh-oh! Fine-tuning LLMs compromises their safety, study finds

Their experiments show that the safety alignment of large language AI models could be significantly undermined when fine-tuned.