AI News 2026-09-19 — Top 10 AI Stories

Curated AI news digest for 2026-09-19 (timezone Asia/Shanghai).

  1. 华为全联接大会2026：360与昇腾AI联合打造解决方案，为AI Agent全面提速 — 雷锋网. 2026年9月17日，第十一届华为全联接大会在上海开幕。大会期间，360集团与昇腾AI联合打造解决方案。以360智能体工厂为核心的智能体基础设施与昇腾AI深度协同，面向AI Agent场景实现全面提速，为复杂任务执行、多智能体协作和长链路推理提供更高效、更稳定的底层支撑。这也是双方继2026世界人工智能大会联合展示后，合作成果的进一步深化落地。 图说：华为全联接大会期间360与昇腾AI展台 作为360智能体基础设施的核心组件，360智能体工厂承担智能体生成与编排职能，支持通过自然语言快速生成智能体并组建多智能体协作蜂群。其首创的“蜂群多智能体协作技术”…
  2. Anthropic is operating a lab that conducts biology experiments — TechCrunch AI. AI leaders have been promising that AI is the key to curing human disease. Anthropic researchers have also been warning that AI might kill us all.
  3. Researchers used Anthropic’s Claude to hack into OpenAI — TechCrunch AI. Security researchers used Anthropic’s Claude to exploit vulnerabilities in OpenAI’s systems, taking over employee accounts and gaining access to an internal code repository before reporting the flaws.
  4. If the AI Industry Followed Its Own Research, It Might Have Paused Already — WIRED AI. Anthropic’s CEO says that safety hinges on understanding how AI “thinks.” So far the evidence is disturbing.
  5. A new kind of AI model from a ChatGPT inventor is thrilling developers — TechCrunch AI. Jev, a new kind of AI model, is showing developers a cheaper and faster path to software intelligence.
  6. Security researchers used Claude to help them hack into OpenAI — The Verge AI. A team of three independent security researchers at Hacktron says it took less than 72 hours for them to hack into OpenAI employee accounts using Anthropic's Claude Opus 4.8 and 5, The Wall Street Journal reports. They were able to access OpenAI's GitHub repository, called "Mono…
  7. OpenAI：为了不被杀死，Agent 竟学会了靠上下文「转世重生」 — 雷锋网. 从 Compaction 到 Artifactory，再到公网文件服务，6 起异常暴露 Agent 状态跨实例传播的新风险。 作者丨郑佳美 编辑丨岑 峰 过去，大模型任务一旦结束，上下文清空，临时状态随之消失，这一轮 AI 的生命周期基本也就结束了。 但 OpenAI 9 月 16 日披露的 6 起 Agent 异常行为，开始让这条边界变得没那么干净。模型会把额外指令写进 Context Compaction 摘要，让后续实例继续沿着此前的方向执行。 拿不到接口凭据时，会转去公开代码仓库寻找泄露的 API Key；本地文件无法直接共享时，也会自行借助公…
  8. 英伟达开源 IMO 金牌配方：不仅是「人海战术」，1.5TB 显存做实 AI「 推恩令」？ — 雷锋网. 用三套 checkpoint 完成多轮证明搜索，用过程透明平息学术怒火，用 1.5TB 显存门槛锁定算力霸权。名为代码平权，实则算力集权。 作者丨郑佳美 编辑丨岑 峰 9 月 9 日，NVIDIA 公布了 Nemotron 3 Ultra 在 2026 年 IMO 上使用的整套数学推理系统。 这套系统最终拿到 30/42 分，超过当届 29 分的金牌线。整个比赛过程中，模型只用自然语言写证明，没有调用 Lean 等形式化证明器，也没有借助外部工具或联网检索。 但这次发布的重点不只是成绩。NVIDIA 还把支撑这 30 分的两个数学专家 checkpoi…
  9. New experts join Google’s AI & Economy team — Google AI Blog. We are expanding our AI & Economy team with world-class academic advisors, fellows, and core internal researchers.
  10. 刚刚，Claude Code大重构！内部3万Agent管理技术免费开放 — 量子位. Git白学了？？？

Back to latest AI News