Anthropic restricts OpenClaw access

Anthropic has shut a door that many Claude power users had quietly been using. As of April 4, Claude Pro and Max subscribers can no longer spend their included subscription usage through OpenClaw, the open-source agent framework that lets people run Claude and other models inside third-party tools. They can still use OpenClaw. They just cannot do it on the flat monthly plans they were already paying for. Anthropic now pushes those users onto a separate pay-as-you-go path called “extra usage.” (pcmag.com) That sounds like a small billing tweak. It is not. It changes what a Claude subscription means. Anthropic’s own help pages say Pro and Max are a unified subscription for Claude’s apps and Claude Code, its official terminal tool. The company also says those plans share one usage pool across Claude and Claude Code. But once a customer steps outside Anthropic’s own interface and into a third-party framework, that shared pool stops counting. The meter starts over. (support.claude.com) OpenClaw mattered because it sat right on that boundary. Released in late 2025, it became one of the more visible open-source ways to turn frontier models into working agents that could execute tasks across tools and services. It supported Claude, ChatGPT, Gemini, Grok, and others. For developers, that flexibility was the point. One interface. Multiple models. Fewer reasons to commit to a single vendor’s workflow. Anthropic’s new rule cuts directly against that logic. (pcmag.com) Anthropic’s public explanation is capacity. Boris Cherny, who leads Claude Code, said the subscriptions “weren’t built for the usage patterns of these third-party tools,” and that the company is prioritizing customers using Anthropic’s own products and API. PCMag reported that the change starts with OpenClaw but is expected to extend to other third-party tools in the coming weeks. That makes this less a dispute with one project than a broader policy decision about who gets to sit on top of Claude’s consumer plans. (pcmag.com) The billing mechanics make the decision easier to understand. Anthropic’s support docs say extra usage is billed at standard API rates, separate from the subscription fee, and requires users to add funds or enable auto-reload. Its API billing docs say those charges run on prepaid usage credits, which can be spent on API access, Workbench, and Claude Code. In other words, Anthropic is steering heavy or unusual workflows away from the soft edges of subscription pricing and back into the harder logic of metered compute. (support.claude.com) That matters because subscriptions and APIs are not just two payment methods. They create different kinds of power. A subscription gives users a sense that they bought access to a product. API billing gives the platform tighter control over routing, limits, and margins. Once third-party agent frameworks have to use metered credits instead of included plan usage, Anthropic gets cleaner economics and more leverage over the ecosystem that formed around Claude. The company can still say the tools are allowed. It just makes the independent path more expensive. The timing sharpens the point. PCMag notes that OpenClaw creator Peter Steinberger joined OpenAI in February, and Sam Altman said OpenAI would support the project as open source. Steinberger said he had pleaded with Anthropic to reconsider and only managed to delay the restriction by a week. So this was not an accidental breakage. It was a deliberate line in the sand, enforced on April 4, with Anthropic’s own extra-usage switch waiting on the other side. (pcmag.com)