Role Confusion 이용한 CoT Forgery 공격으로 Jailbreak 성공률 60% 달성
Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection
Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection
Anthropic’s Fable/Mythos shutdown is the first real model export-control shock
Mythos 모델 오용 방지를 위한 30일 데이터 Retention 정책 도입
What Is AI Red-Teaming? A Practical Introduction for Security Professionals
Your Agentic AI's Safety System Gets Dumber As It Thinks Longer