Furina

New 'Furina' Attack Exploits LLM Safety Instability

Researchers have developed a new attack method called Furina that exploits instability in the safety alignment of large language models. This attack capitalizes on a phenomenon where small input changes can lead to unpr…