Anthropic restricts OpenClaw access
Reports say Anthropic temporarily restricted the creator of the OpenClaw wrapper from accessing Claude models, framing the move around tighter enforcement of third‑party access and policy compliance. The story is sourced to industry reporting rather than an Anthropic press release. (cxodigitalpulse.com)
Anthropic briefly suspended OpenClaw creator Peter Steinberger’s access to Claude on April 10, then restored it hours later. (techcrunch.com) Steinberger posted on X that Anthropic flagged his account for “suspicious” activity while he was testing Claude compatibility for OpenClaw, an open-source wrapper that lets developers run Claude through third-party agent tools. He said the suspension would make it harder to keep OpenClaw working with Anthropic models. (techcrunch.com) The lockout came six days after Anthropic changed how Claude subscriptions work with OpenClaw and similar tools. In an April 4 customer email, Anthropic said Claude subscription limits would no longer cover “third-party harnesses including OpenClaw” starting at noon Pacific that day. (techcrunch.com) A harness is the software layer that keeps an artificial intelligence agent running across repeated tasks, retries, and tool calls, instead of a single chat prompt. Anthropic said those usage patterns were different enough that customers would need pay-as-you-go billing through the Claude application programming interface, not flat-rate subscription access. (anthropic.com, techcrunch.com) Anthropic’s head of Claude Code, Boris Cherny, said on X that the company’s subscriptions “weren’t built for the usage patterns” of third-party tools and that the change was meant to manage growth sustainably. He also said Anthropic was offering full refunds for subscribers affected by the policy shift. (techcrunch.com) Steinberger disputed Anthropic’s framing and tied the move to competition with Anthropic’s own agent products. After the April 4 pricing change, he wrote that Anthropic had “copy[ied] some popular features into their closed harness” and then “lock[ed] out open source.” (techcrunch.com) TechCrunch reported that an Anthropic engineer replied publicly during the April 10 dispute and said Anthropic had “never banned anyone for using OpenClaw,” offering to help. Steinberger later said his account had been reinstated the same day. (techcrunch.com) The episode lands as Anthropic expands its own agent infrastructure. On April 9, Anthropic published a new post describing “Managed Agents,” a hosted service for long-running agents on the Claude Platform, after earlier engineering posts on harness design for autonomous software tasks. (anthropic.com, anthropic.com) Steinberger, who recently joined OpenAI, told TechCrunch he still uses Claude only for testing so OpenClaw updates do not break Claude support for existing users. Anthropic had not published a press release about the suspension as of April 12. (techcrunch.com, anthropic.com)
