Anthropic vs OpenClaw: The Third-Party Harness Lockout of April 2026

On April 4, 2026, Anthropic announced that third-party agent harnesses — most notably OpenClaw — could no longer use Claude subscription plans (Pro/Max) for their API calls. Users would need to pay per-token API rates, with a new "Extra Use" tier offering up to 30% off at $1,000+ monthly spend. ## The Business Problem Claude subscription plans (Pro at $20, Max at $100/$200) subsidize token usage. Power users running third-party harnesses could consume $5,000+ in compute costs on a $200/month plan. As Boris Cherny (Anthropic's head of Claude Code) stated, "subscriptions weren't built for the usage patterns of these third-party tools." Additionally, Anthropic argued that Claude Code and Claude Co-Work are engineered to maximize prompt cache hit rates — reusing previous computation for repeated context. Third-party harnesses often bypassed or underutilized these caching mechanisms, making equivalent workloads dramatically more expensive for Anthropic to serve. ## The "Copy Then Close" Criticism Peter Steinberger, OpenClaw's creator, articulated the core grievance: OpenClaw's open-source community innovated features for AI agent workflows, Anthropic incorporated many of these into Claude Code, and then locked out the open-source community that created the innovations. This "copy then close" pattern — where a platform company benefits from open-source experimentation, absorbs the best ideas into their proprietary product, then restricts the original tools — is a recurring dynamic in the tech industry. ## Community Split The community reaction reflected genuinely valid positions on both sides: **Anthropic's case:** $200/month plans cannot sustain $5,000/month compute costs per user. Unoptimized third-party tools waste compute that could serve other users. The company needs sustainable economics, particularly ahead of a potential Q4 2026 IPO. **The community's case:** Open-source developers evangelized Claude and built its developer reputation. Power users who were Claude's most vocal advocates became its loudest critics overnight. OpenAI explicitly welcomed third-party harness usage on their Codex plans, and Peter Steinberger was acqui-hired by OpenAI — an ironic reversal where OpenAI became the more "open" platform. ## Broader Context The lockout came amid a streak of PR challenges for Anthropic: the Claude Mythos internal document leak, the Claude Code source code leak, and a DMCA enforcement action that removed 8,000+ GitHub repositories. Whether these events collectively damage Anthropic's developer trust long-term or are forgotten within months remains an open question.