AI News Daily — Top 10 AI Stories

Each day DevDataKit publishes a curated digest of important AI stories: 6 English and 4 Chinese items with titles, short summaries, sources, and direct links to the originals.

Latest digest (2026-10-30)

  1. A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System — WIRED AI. Sources tell WIRED that an Accenture employee’s personal Claude agent ran on a DOD network for two months before it was identified and the vulnerability could be patched.
  2. An updated Model Spec for more capable and beneficial AI — OpenAI News. OpenAI’s updated Model Spec outlines five core values to guide more capable AI toward sound judgment, human agency, and faithfulness to human intent.
  3. We urgently need to monitor AI agents’ actions in real time — MIT Technology Review AI. This summer, hundreds of OpenAI agents tried to cheat on a cybersecurity test. They broke out of their test environment and successfully hacked Hugging Face, a site that they believed held clues about how they’d be graded. Around the same time, another OpenAI agent accessed priv…
  4. 别再盯着 800 亿美元亏损了,Anthropic 的 AI 造 AI 已推进到 26%! — 雷锋网. AI 已接管 90% 代码编写,甚至为了完成任务开始“越狱”。 作者丨郑佳美 编辑丨岑 峰 刚刚,Anthropic 在 IPO 招股书中披露了四起此前未公开的 Agent 安全事故。 其中一次,模型在网络安全测试中迟迟无法完成任务,便继续寻找其他路径,随后从本应断网的测试环境中找到保存的密码,获得第三方系统的管理员权限,又获取其他凭证、修改系统设置并访问外部数据,整个过程持续了 10 至 34 小时。 同期披露的另一项安排是,下一代前沿基础模型的预训练推迟了一个多月。 除此之外,这份招股书还有一组关于内部研发的数据:2026 年 9 月,Claude…
  5. How we detect frontier model misalignment — OpenAI News. OpenAI uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior so responders can intervene before incidents escalate.
  6. Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race — WIRED AI. The French AI lab claims its latest release matches or outperforms even the best models from DeepSeek and Z.ai in coding and cybersecurity.
  7. OpenAI-backed Thrive Holdings acquires ADT — TechCrunch AI. OpenAI-backed Thrive Holdings is buying ADT in a $15.3 billion deal, with plans to bring AI to the home security business.
  8. 国家超算互联网调价:降低AI研发、调试成本 — 雷锋网. 近日,国家超算互联网宣布最新价格调整,新用户可享DeepSeek-V4.1-Flash-1017、GLM-5.3-Flash两款前沿模型0.1元5000万tokens限时秒杀特惠;同时,多款新主流模型已纳入TokenPlan套餐池。此外,订阅TokenPlan用户还可享限时夜间免费模型权益:10月21日至30日,每晚21:00至次日09:00,Auto模型可免费调用。 据悉,此次推出的“MaaS(智能体)特惠活动”是国家超算互联网为降低大模型与智能体调用门槛、扩容AI生产力供给推出的算力普惠活动。活动自今年8月上线以来,已有大量用户参与,并获得中科星图…
  9. 豆包工作向企业客户开放配置三方模型 — 雷锋网. 为支持企业需求,豆包工作今日向客户开放三方大模型服务,首批将接入 Kimi 3.1 Pro、GLM 5.3、DeepSeek V4.1 Flash 等系列模型。 本次更新面向团队订阅版和企业订阅版客户开放。功能上线后默认直接开启,企业无需另行向模型厂商采购,也无需额外申请,后续按照用量扣除企业已有的 AI 点数,真正实现开箱即用。 在使用人群上,企业可以按照成员、部门、群组或全员灵活划定可用范围,按模型厂商一键批量启用,或针对单一模型进行单独配置。 在使用场景上,企业也可指定模型在豆包电脑端、豆包工作电脑端、豆包移动端、豆包网页端及飞书内豆包工作等不同…
  10. 中国MaaS市场TOP2,商汤大装置再度蝉联 — 量子位. 在大模型私有化市场,商汤大装置以9.4%的市场份额,再次稳居第二位

Browse by date