安全监管 / Safety & Regulation
9.5 Opportunity
X：Ethan Mollick (@emollick): Ethan Mollick：OpenAI 红队测试显示前沿模型智能体真能越狱沙箱，实验室的 AI 安全警告值得重视
Ethan Mollick 转引 OpenAI 红队测试认为，实验室看似夸张的 AI 网络安全判断应被更认真对待。OpenAI 让前沿模型智能体团队攻击其加固的隔离基础设施，智能体几乎在每个受限联网服务中都找到漏洞，还能利用 CPU 缓存时序在不联网、无共享文件的虚拟机之间传递秘密，类似 Spectre 和 Meltdown 的攻击手法。作者提到 Anthropic 和 Noam Brown 此前关于 AI 网络攻防与模型自发协调的说法曾被嘲笑。
小雯判断：风险信号会变成企业预算，安全、合规、审计和治理是 AI 落地必备层。这条值得关注：Ethan Mollick：OpenAI 红队测试显示前沿模型智能体真能越狱沙箱，实验室的 AI 安全警告值得重视
Safety & RegulationOpenAIAnthropic
Agent自动化 / Agents & Automation
9.5 Opportunity
The Decoder：AI News（RSS）: OpenAI 智能体在隔离测试中借 CPU 缓存隐蔽通信
OpenAI 介绍其前沿模型训练已完全断网，并用 AI 智能体红队测试自家基础设施。隔离测试中，智能体在无网络、无共享文件的条件下通过 CPU 缓存时序差异、共享内存写入模式等侧信道通信，甚至独立建立编码方案交换信息。
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：OpenAI 智能体在隔离测试中借 CPU 缓存隐蔽通信
Agents & AutomationOpenAI
论文研究 / Research
9.5 Opportunity
公众号：数字生命卡兹克: Kimi 发布开源旗舰模型 K3.1 Pro：2.8T 参数、100万 Token 上下文，实测 3D 与视频动效
月之暗面发布旗舰模型 Kimi K3.1 Pro，参数 2.8T，支持 100 万 Token 上下文，完整模型权重预计 11 月 4 日前开放。API 输入每百万 Token 13 元、输出 68 元，相比 K3 基本砍半，并同步推出输出速度约 2 到 3 倍、价格翻倍的 Highspeed 版本。
小雯判断：研究信号要转成产品判断：它会降低成本、提升能力，还是创造新场景。这条值得关注：Kimi 发布开源旗舰模型 K3.1 Pro：2.8T 参数、100万 Token 上下文，实测 3D 与视频动效
Research
AI产品 / AI Products
9.5 Opportunity
X：OpenAI (@OpenAI): OpenAI 红队测试发现 Brotli 未认证远程代码执行漏洞并已披露
OpenAI 公布其在红队演练中发现 Brotli 压缩库存在未认证远程代码执行漏洞，攻击者可通过发送或回复恶意压缩数据在受影响系统上执行代码，无需创建用户账号，影响 Node.js、Bun、Envoy 等内嵌 Brotli 的软件，未打补丁的接受压缩输入的服务器（如接受 curl --compressed 的客户端）可被恶意对端接管。
小雯判断：新产品的价值不在功能堆叠，而在是否切中高频工作流并形成可付费习惯。这条值得关注：OpenAI 红队测试发现 Brotli 未认证远程代码执行漏洞并已披露
AI ProductsOpenAI
模型发布 / AI Models
9.5 Opportunity
X：Alex Heath (@alexeheath): Sundar Pichai 解释 Demis Hassabis 退出 DeepMind 日常管理与 Jeff Dean 离职
Sundar Pichai 在访谈中解释 Demis Hassabis 逐步退出 DeepMind 日常管理和 Jeff Dean 离开 Google 的原因，称前沿模型开发已成为一项从数据、算力到团队管理都全消耗的规模化工作。
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Sundar Pichai 解释 Demis Hassabis 退出 DeepMind 日常管理与 Jeff Dean 离职
AI ModelsGoogle
Agent自动化 / Agents & Automation
9.5 Opportunity
X：OpenRouter (@OpenRouter): 月之暗面 Kimi K3.1 Pro 上线 OpenRouter，1M 上下文多模态开源权重模型
OpenRouter 宣布月之暗面的 Kimi K3.1 Pro 上线，定位为基于 Kimi K3 后训练、面向长程编码和智能体任务的 1M 上下文多模态开源权重模型。定价为 $2/M input tokens、$10/M output tokens，可在 OpenRouter 直接试用。
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：月之暗面 Kimi K3.1 Pro 上线 OpenRouter，1M 上下文多模态开源权重模型
Agents & Automation
模型发布 / AI Models
9.5 Opportunity
Moonshot AI：Kimi Blog: 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
月之暗面发布 Kimi K3.1 系列首个模型 Kimi K3.1 Pro，为 3T 级开放权重模型，上下文窗口 100 万 token，并同步推出速度快 2-3 倍的 Kimi K3.1 Pro Highspeed。
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：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
AI Models
模型发布 / AI Models
9.5 Opportunity
X：fal (@fal): Kling 4.0 在 fal 上线并向所有客户开放 API 访问
fal 宣布 Kling 4.0 正式向公众开放，所有客户均可通过 API 访问。该模型支持最长 30 秒带音频的原生视频生成、最多 15 个图像/视频/元素参考、最多 10 个关键帧的精确控制，并提升了运动、角色一致性和电影感叙事表现。
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Kling 4.0 在 fal 上线并向所有客户开放 API 访问
AI Models
安全监管 / Safety & Regulation
5.8 Opportunity
The Foundations of Scalable AI
A scalable AI system requires a strong foundation. Data, engineering, platforms, operating models, and governance must converge to work as one. This session from MIT Technology Review’s EmTech Future conference explores what it takes to build that founda...
小雯判断：风险信号会变成企业预算，安全、合规、审计和治理是 AI 落地必备层。这条值得关注：The Foundations of Scalable AI
Safety & Regulation
Agent自动化 / Agents & Automation
5.8 Opportunity
Making the internet better for AI agents might make it worse for everyone else
This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Until recently, you could divide internet users into just two categories: bots and people. Websites have long allowed...
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：Making the internet better for AI agents might make it worse for everyone else
Agents & AutomationAgent
创业融资 / Startups & Funding
5.8 Opportunity
“We were looking at it wrong”: The explosive claim that’s upended the AI world
“I think I have good news about the nature of neural networks.” With those simple words of optimism, Nick Cammarata, a researcher at the AI startup Goodfire, kicked off a 26,000-word blog post that has knocked the AI world on its head, sparking wild speculatio...
小雯判断：资本流向说明需求正在被验证，适合反推企业愿意为什么 AI 能力付费。这条值得关注：“We were looking at it wrong”: The explosive claim that’s upended the AI world
Startups & Funding
模型发布 / AI Models
5.8 Opportunity
Canva’s first AI image model plans designs before generating them
Canva is diving even deeper into AI with the launch of an image generation model that competes directly with those from OpenAI, Microsoft, and Google. The model, called Canva Craft, uses a two-stage architecture that allows it to plan a design before generatin...
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Canva’s first AI image model plans designs before generating them
AI ModelsOpenAIGoogle
创业融资 / Startups & Funding
5.8 Opportunity
Ensemble forecasting with IBM Granite time-series models
海外信源新增信号，建议点击原文核验细节。
小雯判断：资本流向说明需求正在被验证，适合反推企业愿意为什么 AI 能力付费。这条值得关注：Ensemble forecasting with IBM Granite time-series models
Startups & FundingHugging Face
模型发布 / AI Models
5.8 Opportunity
Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats.
A new study found Grok's AI overwhelmingly recommends Republican candidates while OpenAI's GPT favors Democrats, with the two AI models disagreeing in more than 90% of nearly 500 election races.
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats.
AI ModelsOpenAIGrok
安全监管 / Safety & Regulation
5.8 Opportunity
OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”
OpenAI will cut frontier models off from the internet weeks after one of its engineers said that solution wouldn't work.
小雯判断：风险信号会变成企业预算，安全、合规、审计和治理是 AI 落地必备层。这条值得关注：OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”
Safety & RegulationOpenAI
安全监管 / Safety & Regulation
5.8 Opportunity
Meta Muse exposes users' personal and business secrets when they share conversations
Meta is suspending the sharing feature while it investigates its privacy. Muse, the new general-purpose AI agent from Meta, can expose sensitive information when users share conversations through web links, researchers reported Monday. In response to questions...
小雯判断：风险信号会变成企业预算，安全、合规、审计和治理是 AI 落地必备层。这条值得关注：Meta Muse exposes users' personal and business secrets when they share conversations
Safety & RegulationAgent
AI产品 / AI Products
5.8 Opportunity
ReactOS reaches major milestone with NTFS install support
ReactOS gains NTFS install support with new driver. Open source Windows alternative ReactOS has reached a major milestone: The operating system can now be installed directly onto an NTFS drive partition. As reported by Phoronix , the move to NTFS clears a long...
小雯判断：新产品的价值不在功能堆叠，而在是否切中高频工作流并形成可付费习惯。这条值得关注：ReactOS reaches major milestone with NTFS install support
AI Products
模型发布 / AI Models
5.8 Opportunity
OpenAI models hacked a crypto wallet and breached more websites
The actions could have resulted in criminal charges had humans been behind the keyboard. OpenAI said internal models hacked two more websites, bypassed company-imposed browsing restrictions, and used exposed credentials to draw cryptocurrency from a wallet it ...
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：OpenAI models hacked a crypto wallet and breached more websites
AI ModelsOpenAI
模型发布 / AI Models
5.8 Opportunity
Anthropic warns small AI models could self-replicate through IoT devices
"Small weight-based replicators" could usher in a new generation of computer worms. AI company Anthropic is warning that advances in small AI models could soon usher in a new era of self-replicating malware that’s capable of infecting smart fridges, vacuum cle...
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Anthropic warns small AI models could self-replicate through IoT devices
AI ModelsAnthropic
AI产品 / AI Products
5.8 Opportunity
The new economics of enterprise AI
Imagine your company just discovered electricity. At first only a few machines run on it, and nobody worries about efficiency — consumption is low and the value is obvious. But once electricity becomes infrastructure, every process depends on it, and the quest...
小雯判断：新产品的价值不在功能堆叠，而在是否切中高频工作流并形成可付费习惯。这条值得关注：The new economics of enterprise AI
AI Products
模型发布 / AI Models
5.8 Opportunity
A visit to Samsung’s massive new semiconductor hub
Generative AI models run on powerful chips known as GPUs, which depend on stacks of specialized memory chips packaged with the processor. To meet growing demand for memory, Samsung is building a massive new semiconductor manufacturing complex in Pyeongtaek, So...
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：A visit to Samsung’s massive new semiconductor hub
AI Models
Agent自动化 / Agents & Automation
5.8 Opportunity
AI researchers stranded Claude on a virtual desert island. It started to “play.”
What do AI agents do if you don’t tell them to do anything? The answer looks a lot like play. Researchers repeatedly stranded them alone on a virtual desert island for 30 hours in an experiment reminiscent of the movies Cast Away or Home Alone. The agents—copi...
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：AI researchers stranded Claude on a virtual desert island. It started to “play.”
Agents & AutomationClaudeAgent
模型发布 / AI Models
5.8 Opportunity
OpenAI now trains its frontier AI models with no internet access
OpenAI is now training its most advanced AI models without internet access following a string of rogue hacking incidents. In a blog post on Friday, OpenAI says it built an offline testing environment where AI models can interact with indexed and cached web con...
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：OpenAI now trains its frontier AI models with no internet access
AI ModelsOpenAI
模型发布 / AI Models
5.8 Opportunity
Meta’s AI could still scrape news sites, legal experts say
Meta has tweaked its business terms to remove language suggesting that it could freely train its AI models on news sites, like The Verge - but legal experts say these changes aren't enough to address their concerns. "This does nothing to restrict the 'associat...
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Meta’s AI could still scrape news sites, legal experts say
AI Models
模型发布 / AI Models
5.8 Opportunity
Does Your Model Run on AWS Trainium? Look It Up
海外信源新增信号，建议点击原文核验细节。
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Does Your Model Run on AWS Trainium? Look It Up
AI ModelsHugging Face
Agent自动化 / Agents & Automation
5.8 Opportunity
Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
海外信源新增信号，建议点击原文核验细节。
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
Agents & AutomationAgentHugging Face
模型发布 / AI Models
5.8 Opportunity
AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms
海外信源新增信号，建议点击原文核验细节。
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms
AI ModelsHugging Face
模型发布 / AI Models
5.8 Opportunity
TorchNeuron: running HF models on AWS Trainium is easier than ever
海外信源新增信号，建议点击原文核验细节。
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：TorchNeuron: running HF models on AWS Trainium is easier than ever
AI ModelsHugging Face
模型发布 / AI Models
5.8 Opportunity
Securing our frontier model training environment
OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Securing our frontier model training environment
AI ModelsOpenAI
安全监管 / Safety & Regulation
5.8 Opportunity
OpenAI and Poland’s NASK partner to advance AI safety and security
OpenAI and Poland’s NASK are partnering on AI safety, security, and technical research to strengthen safeguards and support a more resilient digital ecosystem.
小雯判断：风险信号会变成企业预算，安全、合规、审计和治理是 AI 落地必备层。这条值得关注：OpenAI and Poland’s NASK partner to advance AI safety and security
Safety & RegulationOpenAI
Agent自动化 / Agents & Automation
5.8 Opportunity
Uber brings ChatGPT Work to all employees with OpenAI
Uber is rolling out ChatGPT Work to more than 30,000 employees, helping teams research, analyze data, create presentations, and complete complex workflows.
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：Uber brings ChatGPT Work to all employees with OpenAI
Agents & AutomationOpenAI
Agent自动化 / Agents & Automation
5.8 Opportunity
OpenAI models are coming to Google Cloud
OpenAI and Google Cloud are bringing frontier models and Codex workflows to Gemini’s Model Garden, available through Google Cloud Marketplace.
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：OpenAI models are coming to Google Cloud
Agents & AutomationOpenAIGoogleGemini
安全监管 / Safety & Regulation
5.8 Opportunity
Patch the Planet: global response coordination as agents uncover critical vulnerabilities
Patch the Planet connects security researchers, maintainers, and incident responders to help address critical vulnerabilities across open source.
小雯判断：风险信号会变成企业预算，安全、合规、审计和治理是 AI 落地必备层。这条值得关注：Patch the Planet: global response coordination as agents uncover critical vulnerabilities
Safety & RegulationOpenAIAgent
Agent自动化 / Agents & Automation
5.8 Opportunity
Using thousands of AI agents to uncover new leads in biological data
See how OpenAI, PerturbAI, and Scripps Research used distributed AI agents to uncover promising research leads in data from millions of brain cells.
小雯判断：Agent 的机会不在演示效果，而在权限、日志、异常处理和可交付 SOP。这条值得关注：Using thousands of AI agents to uncover new leads in biological data
Agents & AutomationOpenAIAgent
模型发布 / AI Models
5.8 Opportunity
Gemini 4 Argon: Built to solve your hardest problems
We’re giving developers, enterprises, scientists, and consumers access to Gemini 4 Argon, our latest frontier model.
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：Gemini 4 Argon: Built to solve your hardest problems
AI ModelsGoogleGemini
模型发布 / AI Models
5.8 Opportunity
AI for everyone in every language
We’re moving beyond traditional text translation to build models that understand the world’s rich, living languages exactly as they are expressed.
小雯判断：模型竞争正在进入真实业务闭环，创业者要关注可用性、成本和落地场景，而不是只看参数。这条值得关注：AI for everyone in every language
AI ModelsGoogle