安全 on Joey's AI Labhttps://www.399041.xyz/tags/%E5%AE%89%E5%85%A8/Recent content in 安全 on Joey's AI LabHugozhJoeyFri, 09 Oct 2026 02:17:00 +0800一条提示词劫持 AWS 账户内全部 AgentCore 智能体:AgentCorruption 复盘https://www.399041.xyz/posts/zenity-agentcorruption-aws-agentcore/Fri, 09 Oct 2026 02:17:00 +0800https://www.399041.xyz/posts/zenity-agentcorruption-aws-agentcore/Zenity 披露针对 Amazon Bedrock AgentCore 的攻击链 AgentCorruption:聊天权限即可经 IMDS 窃取凭证,默认角色跨全部智能体,长期记忆可被投毒。Goodfire 给 Kimi K3 和 GLM 5.3 装上生产级网络安全监控:探针 + Judge 级联https://www.399041.xyz/posts/goodfire-production-cyber-monitors/Fri, 09 Oct 2026 02:05:00 +0800https://www.399041.xyz/posts/goodfire-production-cyber-monitors/Goodfire 用激活探针做廉价初筛、LLM judge 只看可疑交互的级联监控:召回 93%、成本降约 50 倍、延迟从 2750ms 到 42ms,已部署进生产。OpenAI 封禁一俄一伊两个隐蔽影响行动,首个 Category 5 出现https://www.399041.xyz/posts/openai-disrupts-dark-clark-bogus-bylines/Fri, 09 Oct 2026 02:02:00 +0800https://www.399041.xyz/posts/openai-disrupts-dark-clark-bogus-bylines/OpenAI 封禁俄罗斯 Dark Clark(首个 Category 5)和伊朗 Bogus Bylines(Category 4)两个假前台影响行动:AI 主要提升规模与语言流畅度,假前台被曝光后依然脆弱。