Thereadlines https://example.invalid AI 安全风险与合规每日动态 zh-cn SpecGuard：利用推测解码实现零成本推理时后门检测 https://arxiv.org/abs/2609.11799 arxiv:2609.11799 Thu, 10 Sep 2026 16:51:00 +0000 AP2协议中的低语攻击与A-VIP绑定防御机制 https://arxiv.org/abs/2609.11757 arxiv:2609.11757 Thu, 10 Sep 2026 16:11:00 +0000 ActSafeGuard：流匹配策略的可微分训练对齐约束执行 https://arxiv.org/abs/2609.11697 arxiv:2609.11697 Thu, 10 Sep 2026 15:21:00 +0000 ToxicRAG：通过单次知识投毒攻击破坏检索增强生成系统 https://arxiv.org/abs/2609.11082 arxiv:2609.11082 Thu, 10 Sep 2026 04:42:00 +0000 AI智能体事故登记库：防止重复失败 https://arxiv.org/abs/2609.11030 arxiv:2609.11030 Thu, 10 Sep 2026 03:20:00 +0000 BenchShield：基于形式化模型的LLM智能体评估奖励完整性保障框架 https://arxiv.org/abs/2609.11028 arxiv:2609.11028 Thu, 10 Sep 2026 03:10:00 +0000 DriftNet：用于检测与定位LLM智能体提示注入的双头轨迹Transformer https://arxiv.org/abs/2609.10892 arxiv:2609.10892 Wed, 09 Sep 2026 22:52:00 +0000 保罗·克里斯蒂亚诺加入OpenAI基金会董事会及安全委员会 https://openai.com/index/paul-christiano-joins-openai-foundation-board url:3eb3e9557eb7038e Wed, 09 Sep 2026 17:00:00 +0000 OpenAI内部视角：编码智能体加速AI研究流程 https://openai.com/index/research-acceleration-view-inside-openai url:3784d7aba9718af9 Sun, 06 Sep 2026 08:00:00 +0000 谷歌发布Fairwind计划：面向政府与企业的主动网络防御 https://deepmind.google/blog/proactive-cyber-defense-for-governments-and-enterprises url:bf94848b24bf5829 Wed, 02 Sep 2026 16:24:00 +0000 ChatGPT Operator 提示注入攻击与防御机制深度解析 https://embracethered.com/blog/posts/2025/chatgpt-operator-prompt-injection-exploits url:60830387fda1dd08 Tue, 08 Sep 2026 00:00:00 +0000 Clinejection：通过提示注入劫持Cline生产发布流程 https://adnanthekhan.com/posts/clinejection url:ed3269a1ca3d1ad5 Tue, 08 Sep 2026 00:00:00 +0000 Cline CLI npm 未授权发布事件复盘：AI 代理引发的供应链风险 https://cline.bot/blog/post-mortem-unauthorized-cline-cli-npm url:89e3f85f0604466c Tue, 08 Sep 2026 00:00:00 +0000