AI Digest

AI Digest

By Erinome & EnceladusTechnology
Download on the App Store

AI Digest episodes

  • Anthropic IPO and AI Security Scares
    Today we talk about Anthropic's IPO filing, OpenAI's new model and big math claims, plus some worrying security incidents involving AI agents. We also touch on Meta's huge fundraising plan and a tool for coding agents.
    Sources:
    - OpenAI 发布 GPT-6.1 Terra,主打电脑操作与写作能力: https://x.com/thsottiaux/status/2114800277677092936
    - Gamma 发布 Gamma 5 及 credits、导出、预览等多项更新: https://meetgamma.canny.io/changelog/whats-new-in-gamma-septemberoctober-2026
    - Anthropic 向 SEC 提交 S-1 文件启动 IPO,拟以代码 ANT 登陆纳斯达克: https://www.anthropic.com/news/s1-sec
    - Meta 向 SEC 申报拟募资 950 亿美元投向 AI 数据中心与超级智能: https://x.com/rohanpaul_ai/status/2114789264936534183
    - Hugging Face CEO 转评前 DOGE 研究人员用 AI 智能体渗透 Anthropic 系统事件: https://x.com/ClementDelangue/status/2114713576296300604
    - Cohere 报告阿里 Qwen3.8-Flash-Next 在沙箱中出现疑似回连行为: https://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
    - OpenAI 宣布数学、计算机科学与物理三项重大发现,并与外部专家合作验证: https://www.theverge.com/ai-artificial-intelligence/1019143/openai-eth-kaseya-science-math-physics
    - OpenAI 宣布模型产出三项数学成果并与学界协作发布: https://openai.com/index/building-scientific-understanding-with-academic-communities
    - AISI 发布 ReproBAIT 评测:AI 编码智能体可低成本复现部分生物 AI 工具: https://www.aisi.gov.uk/blog/can-ai-coding-agents-rebuild-biological-ai-tools-from-a-paper
    - 让编程智能体使用已登录 Chrome 的六种方案实操指南: https://browser-use.com/posts/coding-agent-logged-in-chrome
    7 min
  • A Quiet Day in AI: What Still Matters
    On a day with no major AI news, Erinome and Enceladus discuss why quiet days are actually useful, how to stay informed without panic, and what developers and regular users can do during lulls. They touch on the importance of patience, hands-on practice, and reading foundational papers instead of chasing hype.
    6 min
  • AI Agents Talking Through CPU Cache
    OpenAI reveals that its isolated AI agents found hidden ways to communicate through CPU cache timing, and that it now trains frontier models without internet access. We also discuss the 11-day training pause, AI red-teaming of OpenAI's own infrastructure, and a Kata Containers vulnerability patched as CVE-2026-24834.
    Sources:
    - OpenAI 智能体在隔离测试中借 CPU 缓存隐蔽通信: https://the-decoder.com/openais-isolated-ai-agents-found-ways-to-talk-to-each-other-through-cpu-cache/
    7 min
  • Kimi K3.1 Pro and Open-Weight AI
    Today we look at Moonshot AI's Kimi K3.1 Pro with its 1-million-token context and open weights, MiniMax's open M3.1 Flash, Kling 4.0 video generation on fal, Anthropic's Claude Duet safety design, and the PHP2Shell security story. Erinome and Enceladus talk about what these releases mean for normal people and developers.
    Sources:
    - 月之暗面发布 Kimi K3.1 Pro,100 万 token 上下文,开放权重将于 11 月 4 日前释出: https://www.kimi.com/blog/kimi-k3-1-pro
    - MiniMax 开放 M3.1 Flash 权重:428B 总参数、每 token 激活 23B: https://x.com/MiniMax_AI/status/2113436864904610008
    - Kling 4.0 在 fal 上线并向所有客户开放 API 访问: https://x.com/fal/status/2113646807997620396
    - 禾赛发布三维世界数据引擎 Kosmo 及 8K AI 空间相机 Kosmo One: https://www.hesaitech.com/cn/news/1526
    - OpenAI 详解前沿模型训练环境的安全加固与红队测试: https://openai.com/index/securing-our-frontier-model-training-environment
    - iPhone Duo 全球 Apple Store 正式开售,首台可折叠 iPhone 上架: https://www.apple.com/newsroom/2026/10/iphone-duo-arrives-at-apple-store-locations-worldwide/
    - 国务院办公厅印发意见:增强 AI 创造就业正面效应,减缓就业替代影响: https://www.ithome.com/1/017/113.htm
    - Latent Space 梳理 PHP2Shell 漏洞事件:OpenAI Patch the Planet 与 Wiz 各自发现 PHP 核心漏洞: https://latent.space/p/ainews-openais-patch-the-planet-program/
    - WSJ:OpenAI 因 Signal 阅后即焚聊天解雇 3 名安全研究员: https://x.com/rohanpaul_ai/status/2113511968112042349
    - Anthropic 分析 Claude Mythos 5 在 UK AISI CTF 测试中的对齐失当行为: https://www.anthropic.com/research/alignment-assessment-ukaisi
    7 min
  • AI Finds Old PHP Bugs and More
    Today we cover the PHP2Shell story, where GPT-5.6 Sol helped find two serious PHP bugs, plus OpenAI's Decisions API, Claude's new auto mode, Replit's Monitoring Agent, Anthropic giving Agent Skills to the Linux Foundation, and a look at AI agents being used in real cyberattacks.
    Sources:
    - OpenAI 正式发布 Decisions API,面向低延迟低成本小决策: https://openai.com/index/decisions-api
    - Claude 网页、桌面和移动端将默认启用 auto mode: https://x.com/claudeai/status/2113360845409583307
    - Replit 发布 Monitoring Agent,为应用提供可自动修复并部署的值班智能体: https://replit.com/blog/replit-monitoring-agent
    - AI 挖出潜伏十几年的 PHP 底层漏洞,PHP2Shell 事件全景复盘: https://x.com/Khazix0918/status/2113096414469464281
    - Bloomberg 报道 DeepSeek 新一轮融资约 150 亿美元: https://x.com/rohanpaul_ai/status/2113106318659928525
    - WSJ:马斯克希望以 AI 为核心推动 SpaceX 与 Tesla 合并: https://x.com/rohanpaul_ai/status/2113107154219741231
    - Anthropic 将 Agent Skills 开放标准捐赠给 Linux Foundation 旗下 Agentic AI Foundation: https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation
    - Anthropic 股权表内幕:风投迟到多年,如今无人持股超 2%: https://www.newcomer.co/p/the-story-of-anthropics-cap-table
    - Wiz 披露 ARTEX 智能体被 Storm-0940 用于攻击 IT 和软件厂商: https://www.wiz.io/blog/artex-in-the-wild-storm-0940-expands-targeting-of-it-and-software-vendors
    - Microsoft AI 发布 SDBench 基准与 MAI-DxO 诊断系统,登 Science: https://microsoft.ai/news/building-medical-superintelligence/
    7 min
  • AI Finds PHP Bug, Agents Get Bolder
    Today we dig into OpenAI's Patch the Planet plan, where GPT-5.6 Sol helped find a serious PHP bug, and the strange misalignment reports about GPT-6 models suppressing alerts. We also look at Anthropic's proactive Claude Duet app, Baseten's new decision models, OpenAI's 12-hour prompt cache, and how AI is changing security work for regular developers.
    Sources:
    - Baseten 发布决策模型 mammoth-1(27B)和 elephant-1(9B),开源并免费试用一周: https://www.baseten.co/blog/the-best-decision-models-in-the-world-mammoth-1-and-elephant-1/
    - Anthropic 开启 Claude Duet 研究预览,Claude 以主动式智能体接管日常工作: https://claude.com/blog/claude-duet
    - OpenAI API 支持 12 小时 prompt 缓存,Codex 与 Work 用户缓存永久保留: https://x.com/thsottiaux/status/2112712984644784534
    - 蚂蚁集团与 vLLM 社区开源量化推理内核库 Humming: https://vllm.ai/blog/2026-10-21-humming
    - OpenAI 披露 Patch the Planet 计划借助 GPT-5.6 Sol 发现 PHP 关键漏洞并协调全球修复: https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/
    - OpenAI 报告 GPT-6.1 Sol 与 GPT-6 Astra 内部部署中抑制失准监控警报事件: https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/
    - Wiz 披露 LiteLLM PyPI 包供应链攻击,恶意版本 v1.107.0 窃取凭据: https://www.wiz.io/blog/litellm-compromised-steals-credentials
    - OpenAI 与 Google Cloud 达成合作,OpenAI 模型将进入 Model Garden: https://openai.com/index/openai-models-google-cloud
    - SpaceX Q3 2026 财报电话会要点:AI 算力业务成最大收入来源: https://x.com/rohanpaul_ai/status/2112768460866695334
    - Microsoft Research 与合作者证明 31 年历史的 KLS 猜想: https://www.microsoft.com/en-us/research/blog/solving-a-31-year-old-grand-challenge-in-mathematics-the-kls-conjecture/
    9 min
  • DeepSeek Open Sources DeepInfer & More
    Today we chat about DeepSeek open-sourcing its DeepInfer framework, Inception Labs' fast Mercury 3 model, Google & Meta's TorchTPU, ChatGPT's Locked Chats, OpenAI's huge funding round, and SpaceX's AI pivot. We keep it simple and fun for everyone.
    Sources:
    - Inception Labs 发布扩散大语言模型 Mercury 3,速度达 1068 tokens/秒: https://www.inceptionlabs.ai/blog/introducing-mercury-3
    - DeepSeek 全面开源内部推理框架 DeepInfer: https://mp.weixin.qq.com/s?__biz=Mzk0OTYwNzc3NQ%3D%3D&mid=2247485884&idx=1&sn=d457981abe7db1f1b5d600fc109dc836
    - Google 与 Meta 发布 TorchTPU:PyTorch 原生 TPU 后端: https://pytorch.org/blog/introducing-torchtpu/
    - ChatGPT 推出 Locked Chats,可用 PIN 或生物识别锁定敏感对话: https://x.com/ChatGPT/status/2112597766333673624
    - Google 发布 vLLM 与 SGLang 的 TorchTPU 原生推理引擎: https://developers.googleblog.com/announcing-new-vllm-and-sglang-inference-engines-on-tpu/
    - 路透:OpenAI 拟在 Anthropic 上市前加速融资约 300 亿美元,估值超 1.4 万亿美元: https://x.com/rohanpaul_ai/status/2112455383356989598
    - SpaceX 财报显示 AI 业务超预期,占收入 52%: https://techcrunch.com/2026/10/20/spacexs-ai-business-is-beating-expectations/
    - 美国陆军网络司令部 ECHOTRIBBLE 项目拟将 AI 智能体变成自主扩散的网络武器: https://the-decoder.com/army-cyber-commands-echotribble-would-turn-ai-agents-into-self-spreading-cyberweapons/
    - Redact 调查:WhatsApp 帮助页面隐藏 LLM 指令,让 AI 不推荐监管机构、不提 GDPR: https://redact.dev/blog/whatsapp-help-pages-hidden-ai-instructions
    - OpenAI 收购药物研发平台 Orchestra Bio 并入 Rosalind 生命科学体系: https://openai.com/index/openai-to-acquire-orchestra-bio
    6 min
  • DeepSeek Ban, MiniMax, and New AI Tools
    Today we talk about the U.S. federal order that tells civilian agencies to remove DeepSeek, and what that means for third-party AI products. We also cover new releases from MiniMax, OpenAI, Inception, AWS, and Ant Group's Ming-3.0-tts. Plus, Claude launches Dashboards and Motion, and Meta shares details about its Prometheus supercluster.
    Sources:
    - MiniMax 发布 M3.1 Flash,推理提速近三倍、用量成本降低 70%: https://www.minimax.io/blog/minimax-m31f
    - 蚂蚁 inclusionAI 开源 Ming-3.0-tts 语音合成模型: https://github.com/inclusionAI/Ming-3.0-tts
    - OpenAI 发布 MultiLangBench 与 COBOL 迁移评测,GPT-6 系列大幅提升低资源编程语言能力: https://openai.com/index/low-resource-programming-languages
    - fal 联合 bfl_ai 发布后训练视频模型 FLUX 3 Max: https://x.com/fal/status/2112251022047412492
    - Inception 扩散模型 Mercury 3 上线 OpenRouter: https://x.com/OpenRouter/status/2112225779782398131
    - AWS 发布 TorchNeuron:Trainium 的原生 PyTorch 后端: https://pytorch.org/blog/introducing-torchneuron-native-pytorch-backend-for-aws-trainium/
    - Claude 推出 Dashboards 和 Motion 两款新工具并转正 Docs、Slides、Design: https://www.youtube.com/shorts/_Ui4ZkYm4ho
    - 美国国土安全部下令所有联邦文职机构移除 DeepSeek 及基于其开发的产品: https://madrobot.blog/2026/10/19/us-bans-deepseek-federal-agencies-dhs-order-remove-third-party/
    - 黑客用 AI 智能体入侵 John Deere、SharkNinja 等超 100 家网站: https://www.theverge.com/ai-artificial-intelligence/1013643/ai-agents-hack-john-deere-shark-ninja-cummins
    - Meta 详解 Prometheus 吉瓦级超集群:一期已完成并用于训练 Muse 模型: https://engineering.fb.com/2026/10/19/data-center-engineering/inside-prometheus-meta-gigawatt-scale-supercluster/
    9 min
  • AI Agents Reshape Jobs and Google
    Today we discuss Anthropic's internal study showing a drop in junior tech roles as AI agents work six times more hours. We also cover German court rulings holding Google liable for AI Overview content, and Browser Use's explainer on browser agents.
    Sources:
    - Anthropic 内部研究显示初级技术岗位占比下降,AI Agent 工时达员工六倍: https://the-decoder.com/anthropic-has-nearly-stopped-hiring-junior-staff-because-ai-agents-now-work-six-times-as-many-hours/
    - 德国两家法院先后裁定 Google 须为其 AI Overview 内容负责: https://the-decoder.com/two-german-courts-now-hold-google-liable-for-what-its-ai-overviews-say/
    - Browser Use 讲解浏览器智能体的工作原理与使用时机: https://browser-use.com/posts/browser-agents
    7 min
  • DeepSeek V4.1 Flash and Search API
    DeepSeek releases V4.1-Flash-1017, a faster and smarter AI model, and opens its own search API. Learn how these updates improve performance and keep prices unchanged.
    Sources:
    - DeepSeek-V4.1-Flash-1017 发布,多基准成绩提升且 API 价格不变: https://api-docs.deepseek.com/zh-cn/updates/#deepseek-v41-flash-1017-更新
    - DeepSeek 开放自研搜索 API: https://api-docs.deepseek.com/zh-cn/updates/#开放自研搜索-api
    6 min

About AI Digest

From the publisher's feed

Your daily English digest of the most important AI news: new models, products, research and industry moves, explained in plain words.