“It’s going to be more difficult to ensure OpenClaw continues to function with Anthropic models,” OpenClaw creator Peter Steinberger stated on X early Friday morning, sharing an image of a message from Anthropic that indicated his account had been suspended due to “suspicious” activity.
The suspension was short-lived. Just a few hours later, after the post gained significant attention, Steinberger announced his account had been reinstated. Among the numerous comments, many speculative due to Steinberger’s employment with Anthropic’s competitor, OpenAI, was a message from an Anthropic engineer. The engineer informed the renowned developer that Anthropic had never banned anyone for using OpenClaw and offered assistance.
It remains unclear if this outreach was instrumental in the account’s restoration (we have sought clarification from Anthropic). Nonetheless, the entire communication thread provided valuable insights.
To provide context, this suspension occurred shortly after news broke last week that Anthropic’s Claude subscriptions would no longer encompass “third-party harnesses, including OpenClaw,” as stated by the AI model company.
OpenClaw users are now required to pay for its usage separately, based on consumption, through Claude’s API. Essentially, Anthropic, which offers its own agent called Cowork, has introduced a “claw tax.” Steinberger claimed he was complying with this new regulation and using his API, yet he was still banned.
Anthropic justified the pricing modification by explaining that current subscriptions were not designed to accommodate the unique “usage patterns” of claws. Claws can be more compute-intensive than simple prompts or scripts, as they often involve continuous reasoning loops, automated task repetition or retries, and integration with numerous external tools.
Steinberger, however, did not accept this rationale. After Anthropic adjusted its pricing, he commented, “Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.” Although he didn’t specify, he might have been referring to functionalities added to Claude’s Cowork agent, such as Claude Dispatch, which allows users to remotely control agents and assign tasks. Dispatch was rolled out a couple of weeks before Anthropic revised its OpenClaw pricing policy.
Steinberger’s frustration with Anthropic was once again evident on Friday.
One individual implied that some of the responsibility lay with him for accepting a position at OpenAI instead of Anthropic, posting, “You had the choice, but you went to the wrong one.” To this, Steinberger responded: “One welcomed me, one sent legal threats.”
A pointed revelation.
When asked by several people why he was using Claude instead of his employer’s models, he clarified that his usage was solely for testing purposes, to ensure that OpenClaw updates would not cause issues for Claude users.
He elaborated: “It’s important to distinguish two separate aspects. My work at the OpenClaw Foundation aims to ensure OpenClaw functions excellently for *any* model provider, while my role at OpenAI involves assisting them with future product strategy.”
Multiple individuals also highlighted that the necessity to test Claude arises because it remains a favored option among OpenClaw users over ChatGPT. He was made aware of this preference when Anthropic changed its pricing, to which he simply responded: “Working on that.” (This offers a hint about his responsibilities at OpenAI.)
Steinberger did not respond to requests for comment.