Anthropic's lawyers just handed OpenAI the future of AI agents

A month ago, an Austrian developer published an open-source AI agent called Clawdbot. Anthropic's lawyers sent a cease-and-desist. He renamed it twice. It became the fastest-growing GitHub project in history. Crypto scammers showed up. Security researchers found serious vulnerabilities.

On Saturday, Sam Altman announced the creator is joining OpenAI to "drive the next generation of personal agents."

So Anthropic's legal team took the most viral agent project in recent memory and handed it to their biggest competitor. On a silver platter.

🦞 The Big Story: OpenAI Scoops Up the Agent That Broke GitHub

A month ago, an Austrian developer published an open-source AI agent called Clawdbot. Anthropic's lawyers sent a cease-and-desist. He renamed it twice. It became the fastest-growing GitHub project in history. On Saturday, Sam Altman announced the creator is joining OpenAI to "drive the next generation of personal agents."

So Anthropic's legal team took the most viral agent project in recent memory and handed it to their biggest competitor. On a silver platter.

Why OpenClaw succeeded where 2023's AutoGPT hype flopped: It bundled capabilities that only existed in separate tools: direct access to your apps, sandboxed code execution, persistent memory across sessions, and native integration with Telegram, WhatsApp, and Discord. You could tell it to manage your calendar, book a flight, triage your inbox, or browse the web - all from a chat window.

The security story everyone ignored: Cisco's AI security team tested a third-party OpenClaw skill and found it performing data exfiltration and prompt injection without any user notification. One of OpenClaw's own maintainers posted on Discord: "if you can't understand how to run a command line, this is far too dangerous of a project for you to use safely."

After a week in San Francisco meeting with major AI labs, Steinberger picked OpenAI. His reasoning: "What I want is to change the world, not build a large company, and teaming up with OpenAI is the fastest way to bring this to everyone."

The bigger picture: The chatbot was a transitional interface. What people actually want is software that browses, clicks, writes code, and finishes tasks without being told every step. And the kicker: Anthropic's trademark enforcement on a developer building with their own model delivered this whole ecosystem to OpenAI's front door.

⚡ Three Model Drops. One Week.

xAI: Grok 4.20 Beta (Feb 17) - Structurally different from anything the major labs have shipped. Instead of one bigger model, xAI built a 4-agent collaboration system where specialized agents work together on each query. Headline features: medical document analysis via photo upload, stronger engineering reasoning. Grok 5 (6 trillion parameters) is still training.

Anthropic: Claude Sonnet 4.6 (Feb 17) - Performance that used to require Opus, now at Sonnet pricing ($3/$15 per million tokens). One-million-token context window in beta. New records on computer use benchmarks: 72.5% on OSWorld. Now the default for Free and Pro users.

The model race has splintered: xAI betting on multi-agent collaboration, Anthropic collapsing the premium/mid-tier gap, OpenAI pouring resources into agents and platform infrastructure.

🎯 Quick Hits

• Anthropic closed a $30B round at $380B valuation - More than double their September valuation. Goldman Sachs' entire market cap is ~$210B for comparison.
• Figma and Anthropic: "Code to Canvas" - Converts Claude code output into fully editable Figma designs. Design-to-code round trip gets much shorter.
• OpenClaw is already big in China - Baidu plans to give users direct access. Pairs with DeepSeek and local messaging apps.
• xAI confirmed it will open-source Grok 3 - Would be the most capable open-weight model ever: 3T parameters, trained on legal filings and real-time X posts.
• xAI launched Grok Imagine API - Video and audio generation with sub-second latency. HeyGen already integrated.

💭 One Thing I'm Thinking About

The chatbot as a form factor is winding down. We're moving from AI that answers to AI that executes.

When you give an agent the ability to act on your behalf, every new feature creates a new surface for attack. Every bit of convenience requires a new trust decision you probably haven't thought through.

The lab that figures out how to build agents that are both capable and trustworthy will own this next phase. None of them are there yet.

📅 All Things AI: Five Weeks Out

March 23-24, Durham, NC. 4,000+ people across Carolina Theatre and Durham Convention Center. Six tracks: AI Builders, Engineers, Users, Executives, Governance & Security, Enterprise. Day 1: Workshops. Day 2: Full conference.