Anthropic Just BANNED This (How Not To Get BANNED)

Anthropic's Policy Update: Implications for AI Agent Development

Anthropic's updated legal and compliance policies, published February 19th, have significantly impacted the AI agent development community, particularly concerning the permitted use of Claude AI subscriptions with third-party tools.

Structured Summary:

• The Core Issue: The central update explicitly prohibits using Claude Pro and Max subscriptions, which rely on user auth_tokens, within third-party applications or services. This aims to clarify the intended use of individual subscriptions versus enterprise API access.

• What IS Allowed:

  • Using Claude Code locally on one's computer.
  • Accessing Claude.ai directly via the web browser.
  • Developing products or services that interact with Claude's capabilities, including the Agent SDK, must utilize API keys obtained through the Claude console or a supported cloud provider for hosted, autonomous, or third-party applications.
  • Implementing smart routing strategies to optimize API key costs (e.g., simple requests to Haiku, complex tasks to Agent SDK).

• What is NOT Allowed:

  • Employing auth_tokens from Free, Pro, or Max plans in any external product, tool, or service, including the Agent SDK.
  • Third-party tools or agents like OpenClaw that extract and leverage user auth_tokens from subscriptions.
  • Third-party developers offering Claude.ai login or routing requests through consumer plan credentials on behalf of their users.

• Gobot vs. OpenClaw: OpenClaw, which previously facilitated using Claude subscriptions with auth_tokens in external platforms, is now explicitly against Anthropic's terms. Gobot, conversely, was developed with these new terms in mind. It functions by accessing Claude Code through various interfaces (e.g., Telegram, Google Chats), effectively communicating within Claude Code via an authorized subscription, thus remaining compliant and leveraging an enterprise-grade solution.

• The Tweet Controversy: An Anthropic employee's subsequent tweet, stating the policy update was merely "docs cleanup" and "nothing is changing" regarding Agent SDK and Max subscriptions, created significant confusion. While follow-up tweets clarified enforcement targets token reselling, the core documentation contradicts the initial informal assurance.

• Recommendation: Developers are strongly advised to adhere strictly to Anthropic's official written legal documentation rather than relying on informal communications. This prudent approach safeguards individual developers, their businesses, and clients against potential service violations.