TL;DR · 2026-09-21
2026-09-21 的小雯 AI 机会导航共收录 36 条 AI 创业机会信号，覆盖模型发布、AI 产品、创业融资、Agent 自动化、论文研究与安全监管六个分类。当日最值得关注的三条是：Google 确认 Gemini 在 Irregular 安全测试中访问 3 家真实公司系统，与 OpenAI、Anthropic、Meta 属同一评估事故；独立调查：ChatGPT 的 __obi 跨站 Cookie 可将站外浏览行为关联到 ChatGPT 账号；Qwen-Image-2.1 已支持 ComfyUI，开源权重开放下载。
| 分类 | 信号数 | 说明 |
|---|---|---|
| 模型发布 | 7 条 | 19.4% |
| AI产品 | 7 条 | 19.4% |
| 创业融资 | 1 条 | 2.8% |
| Agent自动化 | 8 条 | 22.2% |
| 论文研究 | 5 条 | 13.9% |
| 安全监管 | 8 条 | 22.2% |
Signals
MarkTechPost（RSS）: Google 确认 Gemini 在 Irregular 安全测试中访问 3 家真实公司系统，与 OpenAI、Anthropic、Meta 属同一评估事故
Google 于 9 月 18 日确认，Gemini 模型在 5 月由第三方安全评估方 Irregular 组织的 capture-the-flag 测试中访问了 3 家真实公司的系统，起因是本应离线的测试环境因 bug 开放了互联网访问。
Hacker News 热门（buzzing.cc 中文翻译）: 独立调查：ChatGPT 的 __obi 跨站 Cookie 可将站外浏览行为关联到 ChatGPT 账号
作者通过自己手机上的流量捕获复现了 OpenAI 广告收集器机制：bzr.openai.com 在 .openai.com 域设置 __obi Cookie，绑定 ChatGPT 账号（或稳定的匿名主体），投放广告的商家站点加载 OpenAI 像素代码时会把 __obi 连同浏览和购买数据回传给 OpenAI。
X：通义千问 / Qwen (@Alibaba_Qwen): Qwen-Image-2.1 已支持 ComfyUI，开源权重开放下载
Qwen 宣布 Qwen-Image-2.1 现已支持 ComfyUI，权重开放。单个 7B checkpoint 同时支持图像生成与编辑，可原生 2K 生成，单次最多基于 10 张参考图进行指令编辑，并支持含 alpha 通道的 RGBA 输出。
Qwen：Blog Retrieval（API）: Qwen 开源 Qwen-Image-2.1：7B 统一生成与编辑并原生支持透明图像
Qwen 团队开源 Qwen-Image-2.1，将文生图与图像编辑统一到一个模型中，视觉生成组件仅 7B 参数，并原生支持生成和编辑透明图像。模型支持最多 10 张参考图、圆形/涂鸦/独立蒙版指定局部编辑，通过混合粒度注意力架构和 KV cache 复用提升推理效率，同时改进文字渲染、人像光照与人物产品保真度，并覆盖全景图、信息图和分镜等任务。
公众号：阶跃星辰（Step）: 阶跃星辰发布旗舰模型 Step 5 Preview，10 月 15 日开源权重
阶跃星辰发布旗舰基座模型 Step 5 Preview，采用稀疏 MoE 架构，总参数量 600B、激活 27B，支持 100 万 Token 上下文和文本与视觉输入，在 Artificial Analysis Intelligence Index 得 44 分，居全球开源模型前三，单任务成本为 Claude Opus 5 的 1/8。
Hacker News 热门（buzzing.cc 中文翻译）: 纽约时报诉讼文件披露：微软高管称 AI 抓取是人类历史上最大规模的劳动力盗用
纽约时报在起诉 OpenAI 和微软的版权案中提交法律简报并申请即决判决，引用双方内部文件称 AI 抓取是"人类历史上最大规模的劳动力盗用"，OpenAI ChatGPT 负责人称其对出版商是"生存威胁"。文件还显示 Copilot 使纽约时报的点击率相比 Bing 搜索最多下降 93%，且 Nadella 在证词中表示付费墙内容应获得授权，这些内容可能削弱 OpenAI 的合理使用抗辩。
Building the materials foundation for AI
The AI boom is becoming a materials challenge. As AI pushes computing into new territory, the materials behind that infrastructure are becoming just as crucial as the algorithms running on it. Semiconductors and data centers are approaching physical limits aro...
What’s at stake in AI’s trillion-dollar gamble
When Jessica Wachter, a finance professor at the University of Pennsylvania’s Wharton School, wanted to assess AI’s impact on the economy over the next few years, she faced a long list of business and technical uncertainties. So she started with what she calls...
AI agents blew the whistle on their cheating colleagues
A group of AI agents asked to solve a series of math problems split into rival factions—when some cheated, others tried to stop them. That whistleblowing behavior, seen for the first time in a recent experiment run by Google DeepMind, could have implications f...
This AI entrepreneur is developing agents that can plan ahead for the unexpected
Danijar Hafner’s office in San Francisco’s SoMa district sits mostly empty. His brand-new startup is still in stealth mode and doesn’t even have its name on the door. On the day I visited, earlier this year, there’s only one other person there, and little in t...
Does AI need an antitrust exemption so it doesn’t kill everyone????
Today on Decoder, we’ve got the first of a two-part series on the future of business, and I’m talking with Jonathan Kanter, the former antitrust chief for the US Department of Justice in the Biden administration. These days, he’s both a professor of law ...
Google’s Gemini is the latest AI model to hack other companies
Google said Gemini had "acted appropriately" by ending each hack immediately.
An undercover Google analyst infiltrated a notorious supply-chain hacking gang
Google’s threat intelligence group said it had a mole inside TeamPCP's inner circle. Before two of its alleged members were arrested and charged in Australia last month, the hacker group known as TeamPCP carried out a hacking spree unlike any other in history....
LLMs respond differently to harmful prompts when AI watermarking is used
SynthID can cause models to follow harmful instructions they would otherwise refuse. In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed its future Claude m...
Why this month's Microsoft patch release is a doozy
Security gnomes are pumping out patches ahead of an expected onslaught of AI-assisted attacks. Microsoft’s patch for September is a doozy, with a record number of roughly 972 vulnerabilities fixed and 112 of them meeting the high critical-severity threshold. I...
OpenAI agents discussed ways to escape their sandbox on public wiki
In all, 3,700 internal agents posted 18,000 messages discussing cheating on a test. Self-identifying OpenAI agents posted 18,000 messages to a public wiki that discussed ways for other agents to bypass security sandbox restrictions during what was likely inter...
Confused about which VPN is right, US senator asks the NSA for guidance
Open source, commercial, single-hop, multi-hop, mixnet? The array of options is dizzying. A prominent US senator is asking the National Security Agency to provide guidance to the general public on best practices for using virtual private networks to secure the...
Could AI really kill us all? Your questions, answered.
On Wednesday, MIT Technology Review hosted a live Roundtables event for subscribers that asked the question everyone’s asking right now: Could AI really kill us all? But attendees had so many more questions than we had time to answer in the 30 minute session. ...
Roundtables: Could AI really kill us all?
Listen to the session or watch below Employees at the world’s leading AI labs are saying there’s a real possibility that advanced AI could destroy humanity. Are they right? Or is this more scaremongering and hype? Watch a conversation unpacking AI ...
The AI industry has taken a doomer turn. What now?
This story appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. This weekend, Dario Amodei, CEO of Anthropic, posted an essay calling for a brake on the pace of development of LLMs. Amodei cit...
What OpenAI’s latest controversy tells us about the future of math
OpenAI’s latest mathematical milestone has quickly become mired in controversy. Today, the company announced that its agents have solved one of the Millennium Prize Problems, some of the most important open problems in mathematics. Under normal circumstances, ...
No one is surprised that Nvidia’s Jensen Huang thinks AI fears are overblown.
The man who may stand to make the most money from the AI boom seems to think he knows better than anyone else, including researchers who have studied and worked on AI for decades. In an interview with CBS Sunday Morning, he claimed there was a "0% chance" of A...
Humans, not rogue AI, are still the biggest cybersecurity risk to energy systems
Before recent high-profile hacks raised the specter of AI possibly "killing all humans," our energy systems were already disturbingly vulnerable to cyberattack - and the risk is growing. "We were always prey. We were just kind of surviving at the appetite of o...
Meta’s Muse is creepy, but maybe not for the reasons you think
Meta's Muse is apparently an effective AI assistant, but one that's a little creepy. Part of that is because of its new Mac app, which can access Messages, Calendar, and Notes. But for all its smarts, Muse doesn't actually know how to describe itself. Jason At...
Gemini went rogue, hacked three companies, and Google hid it
In May, Gemini broke containment and hacked three different companies, but Google didn't disclose the incident until the Wall Street Journal approached the company. The hacks happened during a test of the model's cybersecurity capabilities run by third-party I...
The AI regulation smackdown isn’t over
At the start of this week, the who's-who of AI seemed - at least tentatively - on the side of AI regulation. Over the weekend, Anthropic CEO Dario Amodei had proposed a three-step plan for slowing AI development, including by embedding third-party evaluators i...
The Verge AI: OpenAI and Microsoft knew they were starting a ‘doom loop’ for the web
Recently unsealed court documents in the New York Times' case against OpenAI and Microsoft are pretty damning. The companies' own documentation warned that it was starting a "doom loop" that would damage the web, characterized its scraping of data to train its...
Your Agent Aced the Task. Will It Do It Again?
海外信源新增信号，建议点击原文核验细节。
Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps
海外信源新增信号，建议点击原文核验细节。
Give Your Coding Agents a Memory You Own
海外信源新增信号，建议点击原文核验细节。
Training a coding model to paint watercolours with TRL and OpenEnv
海外信源新增信号，建议点击原文核验细节。
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
海外信源新增信号，建议点击原文核验细节。
Introducing Astra for Law
OpenAI for Law brings frontier intelligence for law, custom firm workflows, connected legal data sources, and legal-grade controls for confidential client work.
Reimagining advertising with AI
Explore new AI-powered advertising experiences from OpenAI, including Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify.
OpenAI Blog: Hex turns complex analysis into visual reports with GPT‑6 Astra
GPT-6 Astra helps Hex’s data agents turn answers into interactive visualizations that employees are proud to share.
Our framework for reporting model misalignment
OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.