Anthropic temporarily banned OpenClaw’s creator from accessing Claude

“Yeah of us, it’s gonna be tougher sooner or later to make sure OpenClaw nonetheless works with Anthropic fashions,” OpenClaw creator Peter Steinberger posted on X early Friday morning, together with a photograph of a message from Anthropic saying his account had been suspended over “suspicious” exercise.

The ban didn’t final lengthy. A number of hours later, after the put up went viral, Steinberger mentioned his account had been reinstated. Amongst a whole lot of feedback — lots of them in conspiracy idea land, on condition that Steinberger is now employed by Anthropic rival OpenAI — was one by an Anthropic engineer. The engineer informed the famed developer that Anthropic has by no means banned anybody for utilizing OpenClaw and provided to assist.

It’s not clear if that was the important thing that restored the account. (We’ve requested Anthropic about it.) However the entire message string was enlightening on many ranges.

To recap the latest historical past: this ban adopted information final week that subscriptions to Anthropic’s Claude would not cowl “third-party harnesses together with OpenClaw,” the AI mannequin firm mentioned.

OpenClaw customers now need to pay for that utilization individually, based mostly on consumption, by means of Claude’s API. In essence, Anthropic, which gives its personal agent Cowork, is now charging a “claw tax.” Steinberger mentioned he was following this new rule and utilizing his API, however was banned anyway.

Anthropic mentioned it instituted the pricing change as a result of subscriptions weren’t constructed to deal with the “utilization patterns” of claws. Claws will be extra compute-intensive than prompts or easy scripts as a result of they could run steady reasoning loops, robotically repeat or retry duties, and tie into a whole lot of different third-party instruments.

Steinberger, nonetheless, wasn’t shopping for that excuse. After Anthropic modified the pricing, he posted, “Humorous how timings match up, first they copy some well-liked options into their closed harness, then they lock out open supply.” Although he didn’t specify, he could have been referring to options added to Claude’s agent Cowork, comparable to Claude Dispatch, which lets customers remotely management brokers and assign duties. Dispatch rolled out a few weeks earlier than Anthropic modified its OpenClaw pricing coverage.

Steinberger’s frustration with Anthropic was once more on show Friday.

One individual implied that a few of that is on him, for taking a job at OpenAI as an alternative of Anthropic, posting “You had the selection, however you went to the flawed one.” To which Steinberger replied: “One welcomed me, one despatched authorized threats.”

Ouch.

When a number of folks requested him why he’s utilizing Claude as an alternative of his employer’s fashions in any respect, he defined that he solely makes use of it for testing, to make sure updates to OpenClaw gained’t break issues for Claude customers.

He defined: “You might want to separate two issues. My work on the OpenClaw Basis the place we wanna make OpenClaw work nice for *any* mannequin supplier, and my job at OpenAI to assist them with future product technique.”

A number of folks additionally identified that the necessity to take a look at Claude is as a result of that mannequin stays a well-liked alternative for OpenClaw customers over ChatGPT. He additionally heard that when Anthropic modified its pricing, to which he replied: “Engaged on that.” (So, that’s a clue about what his job at OpenAI entails.)

Steinberger didn’t reply to a request for remark.