OpenCode 遭网络攻击，防御用的大模型因安全护栏拒绝分析攻击
精选理由
OpenCode 被持续攻击，结果防御用的模型因为护栏拒绝帮忙分析攻击，和 Hugging Face 踩过的坑一样，挺离谱的。
OpenCode 正遭遇持续网络攻击，官方透露事件处理中途遇到难题。用于防御的前沿模型在分析攻击行为时触发了安全护栏，开始拒绝协助对抗攻击代理。这与 Hugging Face 曾遇到的情况相同：安全护栏把攻击分析本身也当作有害内容拦截。该事件暴露了将前沿模型用于网络安全防御时的护栏误伤问题。
原文 · Andrew Curran
OpenCode is currently under a sustained cyberattack, and mid-incident they ran into the same problem Hugging Face did: the frontier models they were using for defense began refusing to help fight attacking agents because analyzing the attacks started triggering safety guardrails. https://t.co/z1YKHVHlhf
