Anthropic blocks OpenClaw’s use based on a subscription quota. Lobster founder: copy the features first, then kick open source out.

Anthropic officially implemented a new policy on April 4, afternoon (U.S. Eastern Time): users on subscription plans such as Claude Pro, Max, etc. will no longer be able to use Claude in third-party AI agent frameworks (harnesses) such as OpenClaw using subscription credits. Users must instead purchase a separate “pay-as-you-go” plan; the fees will be billed independently from their existing subscriptions.

In an email sent to users, Anthropic explained that while this policy starts with OpenClaw, it is “applicable to all third-party frameworks and will be gradually expanded to more tools.” The company also provides a one-time make-good: an amount equivalent to the monthly subscription fee, and it allows users to apply for a full refund.

Lobster founder: copy the features first, then kick out the open source

OpenClaw founder Peter Steinberger publicly expressed strong dissatisfaction with the decision. He posted on X: “What an interesting coincidence—first copy the hot features into your own closed product, then kick the open-source option out.”

Steinberger also revealed that he and OpenClaw board member Dave Morin had tried to persuade Anthropic, and in the end they only managed to get the policy “delayed by one week.” He also made it explicit how to respond: “Bring your own coin (bring your own API Key),” meaning users must obtain and set up a Claude API Key themselves in order to continue using Claude in OpenClaw, without relying on subscription credits.

Anthropic’s explanation: an engineering capacity constraint, not a commercial confrontation

Boris Cherny, the head of Claude Code, responded on X, emphasizing that this policy change is due to engineering capacity considerations rather than targeting the open-source community: “Our subscription plans are not designed for the usage patterns of these third-party tools.” He said Anthropic is trying to manage growth in a planned way to continue serving users well.

Cherny also rebutted Steinberger’s accusations about “copying features,” saying the Claude Code team is “a strong supporter of open source.” He added that he himself had just submitted several pull requests for OpenClaw’s prompt cache efficiency. He said: “This is more of a restriction at the engineering level.”

Background: Steinberger has joined OpenAI; OpenClaw is maintained with support from OpenAI

There is a noteworthy background to this controversy: OpenClaw founder Peter Steinberger joined Anthropic’s rival, OpenAI, in February 2026. After he joined, OpenClaw continued to exist as an open-source project and received support from OpenAI.

VentureBeat’s analysis pointed out that the timing of Anthropic’s policy is quite delicate—OpenAI may be using this opportunity to win over heavy Claude users who are unhappy with Anthropic’s policy, with the image of being “more friendly to open-source developers.”

Actual impact on OpenClaw users

For OpenClaw users, this policy means:

The old way no longer works: after logging into OpenClaw with a Claude Pro/Max account, usage that previously counted toward subscription credits will now be counted as “additional usage” or fail directly

Solution: apply for your own Claude API Key (requires setting up payment in the Anthropic Console), enter it in OpenClaw, and switch to the API billing model

Alternative: switch to models such as GPT-4o, Gemini, local models (Ollama + Llama/Qwen, etc.) that OpenClaw also supports

This controversy reflects a fundamental tension in the AI ecosystem: the way model providers define “fair use” under subscription models, versus the open-source tool community’s demand for seamless access to the strongest models, is forming an increasingly obvious structural conflict.

This article, Anthropic blocks OpenClaw use of subscription credits; lobster founder: copy the features first, then kick out the open source, first appeared on Lianxin ABMedia.