Descriptions:
Anthropic updated its policy documentation to explicitly prohibit using OAuth tokens from Claude Free, Pro, or Max subscription accounts in third-party tools or services — including the Agent SDK. While the policy never mentions OpenClaw by name, the timing is unmistakable: it came weeks after the personal AI agent framework (formerly Claudebot) exploded in popularity, with users routing their $20–$200/month Anthropic subscriptions through it to power persistent personal agents at a significant discount versus direct API pricing.
Matthew Berman, who had upgraded to Anthropic’s $200/month Max plan specifically to use OpenClaw, breaks down what the forced API migration actually costs. Claude Opus 4.6 runs $25 per million output tokens; Claude Sonnet 4.6 runs $15 per million input and $3 per million output. With OpenClaw consuming roughly 50,000 tokens per request at baseline — just for a simple greeting — a single interaction costs approximately $0.25 before any real work begins. Costs spike further when agents read codebases or load large context windows past the 200K token threshold.
Additional clarification from Anthropic’s Tariq partially walked back the Agent SDK restriction, confirming that personal experimentation with Max subscriptions is still permitted, but commercial use requires API keys. Berman frames the entire episode as a significant trust breach with a builder community that was, paradoxically, driving meaningful subscription revenue growth for Anthropic in the weeks prior.
📺 Source: Matthew Berman · Published February 19, 2026
🏷️ Format: News Analysis
