Peter Steinberger, the developer behind the wildly popular open-source artificial intelligence agent OpenClaw, found himself unexpectedly locked out of Anthropic’s systems early Friday morning. Steinberger posted a screenshot on social media showing a message from Anthropic stating that the company had suspended his account due to suspicious activity. He warned his followers that ensuring OpenClaw continues to work properly with Anthropic models will likely become much harder in the near future.
The surprising ban did not last very long. Just 3 hours later, after Steinberger’s post went completely viral across social media platforms, his account was fully reinstated. The comment section under his initial post quickly filled with massive speculation, largely because Steinberger just accepted a high-paying job at Anthropic’s massive rival, OpenAI. One Anthropic engineer even jumped into the messy thread, telling Steinberger that the company never bans anyone simply for using OpenClaw and offering to help sort out the issue immediately. It remains unclear if that specific engineer actually helped restore the account. Still, the entire public exchange highlighted the growing tension between the open-source community and massive corporate AI labs.
This brief suspension follows a major policy shift announced by Anthropic exactly one week ago. The AI company stated that its standard flat-rate subscriptions to the Claude model will no longer cover third-party harnesses, specifically naming OpenClaw in the update. Instead of paying a flat monthly fee, OpenClaw users must now pay for usage separately via a pay-as-you-go API. Because Anthropic also offers its own internal AI agent named Cowork, many independent developers view this new pricing model as a direct “claw tax” designed to punish outside competitors. Steinberger quickly noted he was actually following these strict new rules and using the paid API when the system suddenly banned his account.
Anthropic defended the sudden pricing change by arguing that flat-rate subscriptions simply were not built to handle the heavy daily demands of autonomous agents. Tools like OpenClaw require massive amounts of computing power because they run continuous reasoning loops, automatically retry failed tasks, and connect to over 50 different third-party applications simultaneously. Anthropic claims these intense usage patterns place way too much strain on their server infrastructure compared to a standard user just chatting with a bot.
Steinberger completely rejected that corporate excuse. After the pricing change hit the news, he pointed out the highly suspicious timing of the move. He noted that Anthropic first copied popular agent features directly into their closed software, then immediately locked out open-source alternatives to trap their users. He likely referred to recent software updates to Claude’s Cowork agent, which added premium features like remote task assignment just 14 days before the OpenClaw pricing crackdown officially began.
The deep frustration spilled over again on Friday when a random commenter suggested that Steinberger brought this trouble on himself by choosing to work for OpenAI rather than Anthropic. Steinberger fired back instantly, stating that one company welcomed him with open arms while the other sent aggressive legal threats. This comment references Anthropic’s earlier trademark complaints, which forced him to change his project name multiple times before finally settling on OpenClaw.
When developers asked why an OpenAI employee was even using Claude models in the first place, Steinberger offered a very clear explanation. He stated he only uses Claude for basic testing to ensure that new OpenClaw software updates do not break the experience for the 2 million people who still prefer Anthropic models. He emphasized the absolute need to separate his day job helping OpenAI with product strategy from his volunteer work at the OpenClaw Foundation, where the main goal is to support any model provider the user wants.
Several prominent developers pointed out that Steinberger has to test Claude because it remains the number-one model choice for OpenClaw users, easily beating ChatGPT in daily usage stats. When someone mentioned this strong user preference on Friday, Steinberger simply replied that he was currently working on that exact problem, strongly hinting at his secret new responsibilities at OpenAI.
