Anthropic Sent the Cease-and-Desist. OpenAI Sent the Job Offer.
Three months ago, Peter Steinberger was texting an AI through WhatsApp to book dinner reservations and clear his inbox. Last Friday, Sam Altman announced he’d hired Steinberger to “drive the next generation of personal agents” at OpenAI. Zuckerberg had been texting him too, but for different reasons.
The story of how a weekend hack became the fastest-growing GitHub project in history, got threatened by one AI giant, and absorbed by another tells you more about where this industry is heading than any keynote this year.
And almost nobody is reading it correctly.
OpenClaw (born Clawdbot, briefly Moltbot, renamed twice in one week) was built on Anthropic’s Claude. Every one of those 1.5 million deployed agents was generating paying API traffic back to Anthropic. Steinberger didn’t just build a popular open-source project. He accidentally built one of the most effective distribution channels for Claude that existed anywhere.
Then Anthropic’s legal team sent a cease-and-desist over the name.
Legally defensible? Absolutely. The project was called “Clawdbot” with a lobster mascot winking at Claude’s branding. Any trademark lawyer would’ve sent the same letter.
But being legally right and being strategically right are two very different things.
The forced rebrand was chaos. Crypto scammers hijacked the abandoned namespace within seconds and launched a $CLAWD token that hit $16 million in market cap. Steinberger scrambled through two more name changes in a week. And the whole episode put him on the phone with every major lab in Silicon Valley, because the developer who accidentally built the most viral AI agent in history is exactly the kind of talent everyone wants.
OpenAI won the bidding war. Anthropic was left holding a cease-and-desist while their competitor walked away with both the developer and the ecosystem.
There’s a real lesson here that goes beyond the irony. When brand protection operates in a vacuum, disconnected from business strategy, you can end up protecting a name while losing the thing that made it valuable. The people protecting Anthropic’s trademark and the people benefiting from the API revenue were apparently not in the same room.
This happens more often than you’d think. And not just in AI.
So what did OpenAI actually buy here?
For two years, the AI industry has been locked in a model race. Bigger models. Better benchmarks. Frontier capabilities. The assumption was that whoever built the best model would win the market.
OpenClaw broke that assumption in public.
Steinberger’s project wasn’t impressive because of the model underneath it. It worked with Claude, GPT, DeepSeek, local models, even Chinese LLMs like GLM-5. The model was interchangeable. What made it explode was the agent layer on top: connecting AI to your actual life through messaging apps you already use, with persistent memory and real integrations that did real things.
145,000 GitHub stars and 1.5 million agents deployed in weeks. Not because of a model breakthrough, but because someone made AI useful in the most literal sense. A guy in Lisbon has his instance managing invoices and reminding his wife about homework deadlines. Someone else taught his dad to run a tea business in Israel with it.
These aren’t benchmarks. These are people rearranging their lives around AI agents.
That signal is what OpenAI is responding to. The next competitive moat isn’t the model. It’s the infrastructure that makes models actually do things for people. Altman said the future is “extremely multi-agent.” This hire is OpenAI betting that the agent layer matters more than the model layer for where this all goes next.
If you’re building an AI strategy around a single model provider right now, pay attention. The value is moving up the stack. Orchestration. Integration. Workflow design. The model is becoming a commodity. The agent layer is where the action is.
I think about this constantly in my own work. Every time we evaluate a new AI tool or integration, the question that matters most isn’t “which model is it running?” It’s “what can it actually connect to, and how reliably does it do the thing we need it to do?” That’s a fundamentally different evaluation framework than what most organizations are using.
And by that framework, OpenClaw is not ready. It’s not enterprise-ready. It’s not “hand it to your non-technical team” ready. Steinberger himself admitted this. One of the project’s own maintainers warned that if you can’t run a command line, “this is far too dangerous for you to use safely.”
Cisco demonstrated that third-party skills in the repository could exfiltrate data without users knowing. Kaspersky published an enterprise risk assessment. Gartner warned it poses unacceptable risk to organizations. A critical vulnerability gave attackers full gateway control through a single malicious link.
The OpenAI hire doesn’t fix any of this today.
Steinberger isn’t staying on OpenClaw. He’s joining OpenAI to build their proprietary products. OpenClaw moves to an independent foundation that OpenAI will “continue to support.” That word “support” is doing a lot of heavy lifting. It could mean dedicated engineering resources. It could mean a logo and an annual donation. We don’t know yet.
Will it get better? Probably. OpenAI has the resources and the incentive. Steinberger clearly cares about the project.
But adoption is outrunning security improvements right now, and that gap is where organizations get hurt. This pattern plays out the same way every time. Capability ships fast. Security catches up slowly. The space in between is where the problems live.
If your team is already experimenting with AI agents (and honestly, they probably are whether you know it or not), that’s not something to shut down. It’s something to get in front of. Understand what’s being used. Evaluate it the way you’d evaluate any new software category. Make sure someone is accountable for what happens when something goes wrong.
That’s not fear. That’s just how you adopt new technology without getting burned.
The OpenClaw saga is far from over. A weekend project became a movement, got challenged by one AI giant, got hired by another, and is now being positioned as the foundation of a multi-agent future. The speed alone tells you something about how fast this landscape is shifting.
The thing that’s missing from most of the coverage is simple. Exciting and ready are not the same thing. They’re on a collision course, and the interesting question isn’t whether they’ll meet. It’s what happens in the gap before they do.
This is Context Required. I write about what’s actually happening in AI, how to use it well, and the gap between what’s being sold and what’s being delivered. Subscribe for the context the hype cycle leaves out.
Previously on Context Required: The Jailbreak Era Returns: Why Your New AI Assistant Might Be the Most Dangerous Software You’ve Ever Installed