⚡Megatron LLM Hub
←返回搜索
Twitter

观点:攻击已可一句话完成,防御也应做到一句话可及

把 Agent 安全重心从攻击面清单转向可一键落地的防御链路。
ai_agentllm议题
now that attacking everything is one prompt away a good fulfilling mission for folk, or herein I speak for myself, could/should be: make defending anything one prompt away