Maybe this explains why you got banned?

[JasonCoombs]

Based on the current Claude Code documentation and Terms of Service (as of January 2026), it is generally not a violation to use Claude Code to generate real-time back-end processing for a public live web application, provided you are using the official tools under the correct license.
However, there are significant limitations and risks to be aware of:

1. License and Terms Requirements
   Commercial Use: If you are using Claude Pro or Max, you must ensure your use complies with the Consumer Terms of Service, which may have different limitations than the Commercial Terms used for API/Enterprise clients.
   Third-Party Harnesses Forbidden: You must use the official Anthropic Claude Code CLI or Web interface. Using third-party tools or "harnesses" that hijack a subscription to run Claude Code is a strict violation and has led to bans.
   API Usage for Production: For robust, high-traffic production environments, Anthropic recommends using the Claude API rather than the CLI tool, as the API operates under commercial agreements.
2. Operational Risks (Using AI for Live Backends)
   "Move Fast and Break Things" Risks: Users have reported that relying solely on AI for backend code can lead to critical issues, including:
   Security Vulnerabilities: Claude Code may produce code with SQL injections or misconfigured authentication if not properly guided.
   Data Loss: Using a single production database without proper staging environments, as sometimes suggested by AI, is high-risk.
   Performance Issues: The agent might write inefficient code, leading to poor performance in real-time applications.
   Intellectual Property Risks: There have been reports of Claude potentially ignoring license headers or injecting inappropriate licenses into code, which could cause legal issues in a commercial application.
3. Best Practices to Avoid Violations
   Manually Review Code: Do not push code generated by Claude directly to production without human review.
   Maintain Ownership: Treat Claude as a junior developer rather than an autonomous agent. You must oversee architecture, security, and deployment.
   Follow Official Paths: Use the official Claude Code CLI (claude command) or the official web interface.
   Conclusion: It is allowed, but it is heavily recommended to use it for generating code that you then review, test, and deploy yourself, rather than allowing it to autonomously manage a production-critical backend.

[HugoDaniel]

Thanks for taking the time to go deep on this subject. This is such a detailed response, and makes perfect sense the way that you frame it.

I have since then moved to local LLMs and enjoying it (the "vibe" cli is amazing!), and now I'm not 100% sure if I will be back to claude after this incident. If that happens I'll be more careful.