For a monthly fee of $200, power users of Anthropic’s Claude Max were extracting as much as $5,000 in computational value. This unsustainable math has finally forced the AI safety-first lab to drop the hammer. On April 4, Anthropic announced that its Claude Pro and Max subscriptions will no longer cover usage through third-party ‘harness’ tools, effectively severing the lifeline of popular open-source frameworks like OpenClaw.
OpenClaw, the primary target of this policy shift, was more than just a client; it was a symbol of the 'agentic' frontier. Created by iOS veteran Peter Steinberger entirely through natural language prompts, the tool allowed users to run 24/7 autonomous agents that could manage emails, book flights, and manipulate files. Because it was optimized to minimize telemetry and maximize efficiency, it allowed users to bypass the usage limits inherent in Anthropic’s own web interface while still leaning on the flat-fee subscription model.
The timing of the crackdown is as strategic as it is economic. Over the past six months, Anthropic has systematically introduced internal features—such as Claude Code, Computer Use, and multi-device dispatching—that mirror the core functionalities of OpenClaw. By replicating the 'agent' experience within its own walled garden, Anthropic is pivoting from a model provider to a full-stack platform owner, ensuring that high-intensity compute costs are captured through its own billing cycles rather than subsidized by third-party wrappers.
There is also a layer of corporate intrigue. Peter Steinberger, OpenClaw’s founder, recently joined Anthropic’s arch-rival OpenAI to lead its personal agent efforts. For Anthropic, continuing to subsidize a tool whose creator now works for the competition was likely a bridge too far. This move reflects a broader industry trend where the initial 'open' era of LLM experimentation is being replaced by a defensive, platform-centric 'sovereign' era.
For the global developer community, this marks the end of the 'arbitrage' gold rush. Heavy users who relied on the $200 Max tier to power complex automation workflows now face a choice: pay for high-cost API usage or migrate into Anthropic’s proprietary ecosystem. As Google and OpenAI implement similar restrictions, the industry is sending a clear message: the era of flat-fee, all-you-can-eat compute is over.