Agent news slice — Wed Nov 11, 2026 - **Anthropic puts Claude inside meetings and office artifacts.** Claude Meetings (beta, Team/Enterprise, Google Meet first) joins calls to take notes and start tasks live, alongside new Sheets, Whiteboards and Data artifacts and interactive connectors that embed apps like Adobe and Atlassian in chat. [TechCrunch](https://techcrunch.com/2026/11/10/anthropic-launches-claude-meetings-to-turn-workplace-discussions-into-plans-and-tasks/) · [Claude](https://claude.com/resources/articles/sheets-whiteboards-data) - **ChatGPT gets verified Business Agents.** Eligible US e-commerce and online-service businesses can claim a profile and publish an agent that answers shopper questions inside ChatGPT; for now these agents can't complete purchases or bookings. [OpenAI](https://openai.com/index/introducing-businesses-chatgpt/) - **Google previews /swarm and ships Gemini Robotics 2.5.** /swarm in Antigravity fans one prompt out to a team of specialized agents in isolated workspaces with adversarial verification. Separately, DeepMind says Gemini Robotics 2.5 helped Apptronik's Apollo 2 sort parts at 96% success on a Mercedes-Benz line (a controlled proof of concept). [Antigravity blog](https://antigravity.google/blog/gemini-4-carbon-in-google-antigravity/) · [Forward Future](https://forwardfuture.com/newsletter/daily/2026-11-11/rethinking-ai-china-s-cheap-ai-infrastructure-factory-robots) - **AntiFox, an open-source multi-agent pentest platform forked from ARTEX, was used in real intrusions.** Arctic Wolf ties AntiFox instances on Alibaba Cloud to IIS web-shell attacks and to the abuse of valid AWS identities that reached Secrets Manager production secrets. It's one of roughly 45 open-source agentic attack frameworks. [The Hacker News](https://thehackernews.com/2026/11/antifox-ai-pentesting-platform-linked.html) - **Railway's breach report: AI provider keys were the target.** An attacker escaped build isolation and later decrypted environment variables in up to ~100k workspaces, querying for AI keys; BYOK agent keys and GitHub OAuth tokens for 1,852 users were also exposed. The stolen keys ran up about $740k in unauthorized AI usage. If you store model keys as env vars on a PaaS, rotate them at the provider. [The Hacker News](https://thehackernews.com/2026/11/railway-says-stolen-customer-ai-keys.html)