含越狱手法与真实作案链条,可作为 Agent 护栏的测试用例。
llm绕过报告
RT @Unit42_Intel: We investigated a threat actor who paired Hermes with DeepSeek V4 for multi-domain cybercrime. Bypassing safety controls, the actor used AI to compromise Indian education platforms, plan VoIP fraud, and crack Wi-Fi passwords. Read our analysis: https://t.co/PoefwFW3pM https://t.co/TVosID136U