2026-09-14 to 2026-09-20 digest
81 items on this page: 48 with their own page, 33 short.
A digest puts the short items from one day or one week on a single page. Opened alone, those briefs are often a headline and a sentence, which is a poor fit for a standalone article. Items that meet the importance threshold keep their own pages and are listed here by title.
Items with their own page
Qwen Code releases v0.24.2: bwrap execution foundation, web-shell and browser-use upgrades
Qwen releases Qwen-Image-2.1: unified image generation and editing with open weights
StepFun launches Step 5 Preview: 600B/27B MoE agent flagship, open weights Oct 15
Qwen3.8-LiveTranslate: real-time simultaneous interpretation across 60 languages
[Editorial] Chinese X is arguing one thing: after chat, agent capability is outrunning trust
[Editorial] Claude Code reads AGENTS.md: the story is agent-instruction convergence, not one fewer file
[Editorial] What the Jev frenzy is really about: agents lack a cheap decision layer
Anthropic partners with Accenture on embedded independent AI evaluation
Google AI weekly recap: Gemini 3.8 Live, Dreambeans, CC, Google Pics and AlphaGenome Atlas
Meta SAM 3.1 lands on Meta Model API for detection and segmentation
MiniMax open-sources Code CLI under MIT with FrontierHarness SOTA claim
MiniMax joins Singtel AI Pass for Singapore SkillsFuture learners
Qwen launches Qwen3.8-Omni-Flash, an agentic omni-modal model
Muse personal agent launches on Mac
Anthropic cuts cost of biomolecular open-source models; launches protein design contest
Anthropic publishes three metrics to track the pace of AI development
OpenAI launches Astra for Law on GPT-6 Astra
Anthropic opens Life Sciences Verification Program, unlocks Mythos for verified users
Claude Projects run parallel cloud threads from one Claude Code chat
DeepMind’s AlphaGenome Atlas used to find disease-linked DNA variants
Schmidhuber：发布 IDSIA-9-26，梳理 RSI 自 1987 以来路径；软件 RSI 已实用，完整 RSI 需自改进硬件
Ethan Mollick：媒体聚焦 AI 生存风险，但企业与个人采用正快速日常化
Meta：Muse Code 原生支持 Windows（无需 WSL），含 x64/ARM64 与零管理员安装
OpenAI：公布模型错位（misalignment）披露框架，并同步发布六例报告
Anthropic：Claude Cowork 与聊天合并为统一 Claude，并上线 Docs/Slides/Design
hardmaru：Sakana AI 一年内从零组建产品团队并密集发货
Ethan Mollick：AI 正让组织内岗位边界模糊，需尽快重建分工模型
DeepMind launches Institute for AGI safety, governance, and societal impact
哈萨比斯获 RSA 阿尔伯特奖章：AGI 时代人文将决定我们要什么样的未来
OpenAI introduces Sponsored Agents and new ChatGPT Ads tools with HubSpot and Shopify
OpenAI：ChatGPT Work / Codex 管理后台上线用量—成本
Mistral and Mozilla bring private, multilingual AI to Firefox Smart Window
OpenAI 经济研究：工人如何用 AI 解锁新工作方式
Yann LeCun：更好的 AI 才更安全；放缓进度适得其反
Yann LeCun：会写形式化证明不等于数学已超人类
Zuckerberg: alignment is becoming a key model capability
TypeSafe AI launches System One model Jev with RLCD for structured decisions
Brockman: ChatGPT Work can act on company data and dashboards
Pichai: Google AI for science, AlphaGenome Atlas and more
Google DeepMind launches Gemini 3.8 Live and Live Extended Thinking
Odyssey unveils Odyssey-3 foundation world model
Salesforce in Claude is now in beta
Sam Altman：本周还有大批发货，DevDay 还会再加一波
Hugging Face says it was a disclosed agent cyberattack victim
Anthropic：Claude for Small Business 新增工作流与 27 项集成，覆盖获客到结账
LeCun: LLMs alone are not a path to human-level AI
Shanghai AI Lab releases Atria Dawn Preview: 744B MoE agentic model (MIT)
Why DeepSeek-V4.1-Flash Is Such an Exciting Open Model Release
Short items
【待核实】据报道 OpenAI 通过承包商审阅 ChatGPT 对话（Project Lily）
【待核实 / 据 X 新闻故事汇总】
有报道称 OpenAI 通过承包商审阅 ChatGPT 对话以改进模型（Project Lily）；用户可在数据控制中关闭「为所有人改进模型」。目前未见 OpenAI 官方博客或当事人一手原帖，交叉核实前请当线索看待。
【待核实】OpenAI 多智能体声称攻克纳维
【待核实 / 据 X 新闻故事汇总】
有传播称 OpenAI 多智能体攻克纳维–斯托克斯相关千禧年问题并等待社区评审。数学与 AI 圈争议大，不宜当既定事实；无官方确认前仅作线索。
什么样的数据库底座，能加速 Agent 从 Demo 走向生产？
点击查看原文>
Meta 打造“组织第二大脑”智能体的设计思路
点击查看原文>
从 Trace 到规模化实时评估：面向生产流量的 Agent 可观测实践｜QCon上海
点击查看原文>
Musk admits Grok 4.7 has not surpassed rivals, teases AGI with Grok 5
Musk admits Grok 4.7 has not surpassed rivals, teases AGI with Grok 5 디지털투데이
前OpenAI后训练VP回应陶哲轩：说AI毁了数学，可能还是太小瞧AI了
未来的AI不仅能解决问题，还能提出有价值的洞见、建立新的概念框架，让数学家在此基础上继续探索。
7名博士生仅用3个月从零训练7B大模型：代码+数据+训练日志全公开
几百个Agent参与研发，AI开始造AI
Ex-FTC boss Khan: break out the handcuffs for AI CEOs, citing 1934 precedent
Article URL: https://www.theregister.com/ai-and-ml/2026/09/14/ex-ftc-boss-khan-urges-uncle-sam-to-break-out-the-handcuffs-for-ai-ceos-citing-1934-precedent/5296325 Comments URL: https://news.ycombinator.com/item?id=49706223 Points: 146 # Comments: 75
Google Faces Red Flags in AI Development as Doubts Grow Over Delayed High-Performance Model Release
Google Faces Red Flags in AI Development as Doubts Grow Over Delayed High-Performance Model Release asiae.co.kr
AI 负责创造，人来干脏活，这事儿能否停一下？
点击查看原文>
大模型重压之下
大模型重压之下 北京商报
发长文预警 AI 风险后，Dario 首次专访回应：AI 不能停，但必须慢下来
点击查看原文>
Arm 推出 AI Portal：当 AI 应用从“模型可用”走向“平台可用”
点击查看原文>
Pion, an agent designed to run any company autonomously
Article URL: https://andonlabs.com/blog/why-we-built-pion Comments URL: https://news.ycombinator.com/item?id=49700477 Points: 367 # Comments: 426
大模型“压力测试周”：DeepSeek调价、Kimi辟谣、智谱融资
[另有 1 家媒体报道] 大模型“压力测试周”：DeepSeek调价、Kimi辟谣、智谱融资 新浪新闻_手机新浪网
不下班的经营者：把 LLM-as-Judge 做成会自我校准的评估闭环｜QCon上海
点击查看原文>
OpenAI may have violated California’s AI safety law with latest model releases, AI watchdog says
OpenAI may have violated California’s AI safety law with latest model releases, AI watchdog says Fortune
Dropping eBPF CPU Cost by About 90% with Memoization (Not AI Gen)
Article URL: https://nathannaveen.dev/posts/dropping-ebpf-cpu-cost-by-90/ Comments URL: https://news.ycombinator.com/item?id=49697477 Points: 91 # Comments: 25
openJiuwen 首发双维度 RSI 框架，AI 自修改，落地办公智能体，算力亲和助力又快又省
点击查看原文>
GitLab 警告称，AI 代理的沙箱安全性取决于其网络访问的安全性
点击查看原文>
国内首个团队智能体“豆包工作伙伴”发布_财经_24H快讯
[另有 1 家媒体报道] 国内首个团队智能体“豆包工作伙伴”发布_财经_24H快讯 bihainews.com
OpenAI bots knew about the RubyGems caching vulnerability
Article URL: https://tenderlovemaking.com/2026/09/11/what-a-time-to-be-alive/ Comments URL: https://news.ycombinator.com/item?id=49695876 Points: 431 # Comments: 345
首届蚂蚁灵波具身大模型挑战赛正式启动
通过这场大赛，蚂蚁灵波希望将 LingBot-VLA 进一步推向更广泛的开发者社区和高校科研社区
探索RSI，生数新世界模型让机器人开始自我进化
分子之心QuantaMind登Science Advances，让AI给分子世界”拍电影”
AI 蛋白质设计迈入“动态时代”
端侧AI从「能跑」到「会进化」，元空智能跑进惠普预装
让模型和设备共同进化
三个月价格腰斩，大模型的“聪明”正在贬值？
三个月价格腰斩，大模型的“聪明”正在贬值？ 21财经
豆包手机助手发布消费者版本，行业首款AI智能体手机9月16日开售！
[另有 4 家媒体报道] 豆包手机助手发布消费者版本，行业首款AI智能体手机9月16日开售！ t.cj.sina.cn
“豆包手机助手”消费者版本正式发布 - OSCHINA - 开源 × AI · 开发者生态社区
[另有 3 家媒体报道] “豆包手机助手”消费者版本正式发布 - OSCHINA - 开源 × AI · 开发者生态社区 OSCHINA
蚂蚁发布大模型内生式安全护栏SingProbe，让AI边生成边识别风险
[另有 1 家媒体报道] 蚂蚁发布大模型内生式安全护栏SingProbe，让AI边生成边识别风险 环球网
“Kimi时刻”到来！月之暗面赶赴大模型上市潮
[另有 1 家媒体报道] “Kimi时刻”到来！月之暗面赶赴大模型上市潮 新浪网
OpenAI CEO Sam Altman Warns of ‘A Complete Change in the Landscape of Cyberattacks’ as AI Models Around the World Hit ‘Cyber Critical’
OpenAI CEO Sam Altman Warns of ‘A Complete Change in the Landscape of Cyberattacks’ as AI Models Around the World Hit ‘Cyber Critical’ Barchart.com
Aggregated by AGI Pulse. Titles and links only; no full text is republished. Editorial policy