Anthropic Banned OpenClaw's Creator From Claude. Then Reinstate Him.
Peter Steinberger - the creator of OpenClaw, now employed at OpenAI - was briefly locked out of his Anthropic account. The timing with Anthropic's recent OpenClaw pricing change makes it look like more than a clerical error.
Last Friday, Peter Steinberger - creator of OpenClaw, now working at OpenAI - posted on X that Anthropic had suspended his Claude account. “Yeah folks, it’s gonna be harder in the future to ensure OpenClaw still works with Anthropic models,” he wrote.
A few hours later, after the post went viral, the account was reinstated. An Anthropic engineer showed up in the replies to say they’d never banned anyone for using OpenClaw and offered to help. Which is, at minimum, an interesting PR response to a situation that probably shouldn’t have happened.
The “claw tax” context
This ban didn’t happen in a vacuum. The week before, Anthropic had announced that subscription access to Claude would no longer cover “third-party harnesses including OpenClaw.” Usage through OpenClaw would now need to be paid for separately through the API, based on consumption.
Anthropic’s explanation was that the subscription model wasn’t built for the usage patterns that OpenClaw generates. Claws can run continuous reasoning loops, automatically repeat or retry tasks, and tie into a lot of other third-party tools. That’s computationally expensive in ways a standard Claude subscription wasn’t priced for.
Steinberger’s response: he was following the new rules and using the API. And then got banned anyway.
The OpenAI factor
The story gets more pointed when you know the employment context. Steinberger is now at OpenAI. His take on the situation, posted on X: “One welcomed me, one sent legal threats.”
That framing has stuck. The idea that one AI lab welcomes you and one threatens you - in the context of a developer building an open-source tool that works with multiple model providers - is a meaningful narrative. OpenClaw is explicitly built to be model-agnostic. It works with Claude, it works with GPT, it works with whatever else. And now the lab whose model it was most closely associated with is making it more expensive to use.
Anthropic’s engineer response in the thread was that they’d never banned anyone for using OpenClaw. Which is probably true - this was probably a mistaken automated ban triggered by the new API usage patterns from an account that was suddenly generating different traffic. But the optics matter when you’re a developer who builds tools for multiple AI providers and one of those providers employs you.
What OpenClaw users need to know
If you’re using OpenClaw with Claude, the practical reality is:
- Subscription Claude no longer covers OpenClaw usage
- You need to pay for that usage via the API
- The economics of using OpenClaw with Claude have changed
Steinberger’s follow-up comment: “Working on that.” Given he’s at OpenAI now working on future product strategy, the subtext is probably that OpenAI is thinking about how to make their models work better with OpenClaw. Which would be an interesting competitive move.
The bigger picture
There’s a real tension building between Anthropic’s desire to own the agent experience (Cowork, Claude Dispatch) and the open-source developer ecosystem that built tools like OpenClaw. These tools were partly responsible for Claude’s popularity with developers in the first place. Charging a “claw tax” and then accidentally banning the creator of the most prominent claw tool is not a great look.
Whether this was an algorithmic error, a billing quirk, or something more deliberate, the outcome is the same: one of the most respected independent AI tool developers in the community had a bad experience with the lab he’s now competing with.
Source: TechCrunch