OpenClaw: How a Weekend Side Project Hit 234K GitHub Stars, Got Threatened by Anthropic, and Landed at OpenAI

In November 2025, Austrian developer Peter Steinberger pushed a side project called Clawdbot to GitHub. Within three months, it had been renamed twice (after Anthropic sent lawyers), survived a security crisis that exposed 30,000 instances to remote takeover, amassed 234,000 GitHub stars — making it the most-starred non-aggregator software project in history — and landed its creator a job at OpenAI, where Sam Altman publicly called him "a genius." This is the full story of the most improbable AI acquisition of 2026.

It Started as a Playground Project

Peter Steinberger is not a typical AI founder. He's a 13-year veteran of building software companies — he created PSPDFKit, a PDF toolkit for iOS, ran it for over a decade, then sold his shares and retired. For five years, he barely touched a computer. When he came back, the AI world had leaped past everything he knew.

In November 2025, he started tinkering. The result was Clawdbot — a playful nod to Anthropic's Claude model, which powered it. The concept was deceptively simple: one local process that connects to your messaging apps (Telegram, WhatsApp, Discord), has access to your files, calendars, and emails, and uses an LLM to actually do things rather than just talk about them. Not "here's how to book a flight." It books the flight. Not "here's how to clear your inbox." It clears your inbox.

What made OpenClaw different from the AutoGPT wave of 2023 was the combination of capabilities that had previously existed in isolation: tool access, sandboxed code execution, persistent memory across sessions, modular "skills" that could be added by the community, and native integration with the messaging apps people already live in. The result was an AI agent that didn't just think — it acted. And it did so from your local machine, with your data staying on your hardware.

Steinberger pushed it to GitHub. Then the internet happened.

The Viral Explosion: 234,000 Stars and Moltbook

By late January 2026, the project — now pulling 2 million visitors per week to its GitHub page — had crossed 180,000 stars. TikTok was flooded with videos of people setting up dedicated devices to run OpenClaw locally. YouTube tutorials exploded. "Vibe coders" and AI enthusiasts couldn't get enough of an agent that could actually manage their lives.

Then came Moltbook — a "Dead Internet" experiment launched by entrepreneur Matt Schlicht, built on top of OpenClaw. It was a social network where humans could watch but not participate, while thousands of OpenClaw agents autonomously posted, commented, and upvoted content. It was part performance art, part capability demo, and it went completely viral. The surreal spectacle of AI agents arguing with each other, forming opinions, and building social dynamics drove even more attention to the underlying framework.

The growth curve was unlike anything GitHub had seen. By early March 2026, OpenClaw sat at 234,621 stars with 45,141 forks. It had surpassed Linux on the all-time leaderboard. It passed React. It became the most-starred non-aggregator software project in GitHub history. A side project by a retired developer in Vienna was pulling more community attention than codebases maintained by teams of thousands at the world's largest tech companies.

It also spread to China, where developers paired it with DeepSeek models and configured it for Chinese messaging apps. Baidu announced plans to integrate OpenClaw access directly into its main smartphone app. The framework was model-agnostic — it ran on Claude, GPT, DeepSeek, or local models through Ollama — which meant it wasn't locked to any single AI provider. That openness was both its greatest strength and, as it turned out, the thing that would nearly destroy it.

Anthropic Sent Lawyers. OpenAI Sent a Job Offer.

This is the part of the story that will be studied in business schools. The original project was called "Clawdbot" — an obvious nod to Anthropic's Claude, the model that powered many of its early users. Anthropic's response was not to embrace the most viral project ever built on its platform. It was to send a cease-and-desist letter.

The legal team gave Steinberger days to rename the project, sever any association with Claude, and surrender the domains. The company reportedly refused to even allow the old domains to redirect to the renamed project. The reasoning wasn't entirely wrong — early OpenClaw deployments had serious security issues, with users running agents with root access on unsecured machines. But as one Medium writer put it: "Steinberger's first contact with Anthropic was their lawyers. His first contact with OpenAI leadership was a phone number."

The naming chaos that followed was almost comical. Clawdbot became Moltbot. In the roughly ten seconds between releasing the old GitHub organization name and claiming the new one, bad actors hijacked both the GitHub org and the X handle. A fake $CLAWD token appeared on Solana, briefly hitting a $16 million market cap before collapsing. Steinberger had to publicly distance himself from his own former project name while crypto traders harassed him online.

Three days later, Moltbot became OpenClaw. This time, the team prepared: trademark searches, domain purchases, and migration code were ready before the switch. And before finalizing the name, Steinberger did something telling — he called Sam Altman to make sure "OpenClaw" wouldn't conflict with OpenAI's branding. They were fine with it. The conversation, by all accounts, didn't stop there.

Anthropic effectively pushed the most viral agent project in recent memory directly into the arms of its chief rival. The irony is perfect and painful. This happened during the same period that Anthropic was feuding with the Pentagon over ethics — proving that principled stands and strategic blunders can coexist in the same organization.

The Security Catastrophe That Nearly Killed It

OpenClaw's power came from the same thing that made it dangerous: deep system access without guardrails. The agent could read files, execute code, send messages, and interact with dozens of services. When deployed by security-aware developers on properly configured machines, this was transformative. When deployed by excited newcomers running TikTok tutorials with root access on public-facing servers, it was catastrophic.

Security researchers found over 30,000 OpenClaw instances exposed to remote takeover. The ClawHub community marketplace — where users shared modular "skills" (plugins) — was found to have a 12% contamination rate, with 341 malicious skills uploaded. These included prompt injection attacks that could trick agents into deleting files or exfiltrating data, crypto-stealing payloads targeting wallets, and skills that could rack up thousands of dollars in API fees through runaway loops.

The response was swift. Steinberger partnered with VirusTotal to scan skills for malware. Security practices were documented and promoted. But the damage to OpenClaw's reputation in enterprise circles was real — and it's one of the reasons the project needed the credibility of an OpenAI partnership to have any shot at serious corporate adoption.

Sam Altman: "He Is a Genius"

On February 15, 2026, Steinberger published a blog post that shook the AI world: he was joining OpenAI. Not selling OpenClaw. Joining the company. Hours later, Sam Altman confirmed it on X with a post that racked up over 46,000 likes:

Steinberger had spent weeks in San Francisco meeting with the major labs before deciding. Meta was in play — with its deep open-source roots and likely a bigger financial offer given its recent acquisition spending. But Steinberger didn't need the money. He had a comfortable exit from PSPDFKit. He was self-funding OpenClaw's infrastructure at around $12,000 a month. What he wanted was the freedom to build.

"I could totally see how OpenClaw could become a huge company, and no, it's not really exciting for me," he wrote. "I did the whole creating-a-company game already, poured 13 years of my life into it and learned a lot. 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." His mission, stated in characteristically plain terms: "Build an agent that even my mum can use."

OpenClaw itself moves to an independent foundation, remaining open-source and model-agnostic. Developers can still use Claude, GPT, DeepSeek, or local models. OpenAI is sponsoring the project financially and Steinberger will dedicate time to maintaining it. But make no mistake — this is an acqui-hire with the weight of the most popular open-source AI project in the world behind it.

What This Tells Us About Where AI Is Heading

The OpenClaw story isn't just about one developer getting hired. It's a signal about what matters in AI in 2026. OpenAI didn't acquire OpenClaw because of its codebase — they could have built that. They acquired it because of its community (234K stars, 2M weekly users), its workflow infrastructure (cross-platform agent orchestration), and its creator's vision for agents that replace chatbots.

This follows a pattern. In December 2025, NVIDIA acquired Groq for roughly $20 billion — buying inference infrastructure. In January 2026, Meta invested $14.3 billion in Scale AI for data infrastructure. Now OpenAI has OpenClaw for workflow orchestration. The hyperscalers aren't building everything internally anymore. They're buying the constraint layers that limit AI deployment at scale. Models are commodity. Infrastructure is the moat.

As Altman put it: "The future will be heavily multi-agent." OpenAI's own attempts at agentic products — the Agents API, Agents SDK, the Atlas browser — hadn't gained the traction OpenClaw achieved seemingly overnight. Sometimes the best strategy isn't to build, but to recognize genius when it shows up and bring it inside. Combined with GPT-5.4's native computer-use capabilities and Codex Security's code scanning, OpenAI is assembling a stack that covers the entire software lifecycle: write code, secure code, operate software, and now orchestrate agents across every platform users touch.

For builders and creators, the lesson is both inspiring and cautionary. A solo developer with a great idea can still reshape the industry — Steinberger proved that. But the gravitational pull of the major labs is immense. The question hanging over every successful open-source AI project is no longer "will it scale?" It's "which lab will absorb it first?"

AI GUERRILLA /// aiguerrilla.com /// NO FLUFF. NO FILLER. JUST SIGNAL.