Anthropic 用 Claude 为 123 个开源库的 239 个入口点生成证明。上线评审仍要核对证明命题、部署工件和残余风险，避免把有边界的证明误读为整库无漏洞。
Anthropic's 123-library experiment shows that LLMs can scale proof writing. A deployment assurance manifest shows what was proved, what ships, and wha
WhatsApp 隐藏指令事件说明，官方来源的事实可信度与指令权限必须分开，RAG 系统需要审计模型实际看到的内容。
WhatsApp's hidden chatbot instructions show why RAG systems must separate source trust from instruction authority and audit model-visible content.
SynthID 的规模化能力取决于验证端：版本路由、不确定态、审计证据、人工复核和每个可辩护结论的真实成本。
SynthID scale depends on verification throughput: version routing, uncertainty, audit evidence, review queues, and cost per defensible verdict.
Mercor 估计美国九成工作时间在现有基准中缺少覆盖。本文拆开任务覆盖、模型能力与岗位暴露，给出企业任务地图方法。
Mercor estimates 90% of US work hours lack benchmark coverage. Separate task coverage, model capability, and job exposure before forecasting.
A study of 17,211 citations shows how source preference can amplify ordinary web publishing into an AI search security path.
Third-party AI assessment needs enforceable contracts for independence, access, and disclosure, not a credibility label.
