Will we get banned from Anthropic when using max account?

[Bumblebiber]

Because that's apparently the case when you use your max account for Openclaw.

[SteliosKornelakis]

I have the same question here? Are we using claude code CLI via non interactive flag or we inject the api token in an TOC's breach way?

[saneshark]

Not necessarily. First of all there is still some ambiguity around what this all means and conflicting information around the ToS and a tweet by core developer on X. A number of folks allege getting banned, but I have yet to see a concrete example where that is truly the only plausible justification of the ban.

I opened this issue, now closed, which also highlights how AgentSDK is currently used.

Given the nature of this project and that some of these hooks are likely available within the CLI directly, its possible to move away from AgentSDK use altogether, though as mentioned before and why the issue was closed, probably best to wait for Anthropic to meaningfully address ambiguity in language of ToS against the X post. AgentSDK is a lot cleaner and extensible for building custom skills that depend on hooks and ideally one shouldn't have to create their own abstraction of an agent for the CLI, especially if they are not commercial use.

[wmalgadey]

I didn’t find an official announcement but an explanation of what can be found online.

• https://www.engineerscodex.com/anthropic-claude-subscription-switcharoo

mybe it helps to get a clearer picture in the meantime?

[pratikrath126]

Using personal accounts for high-volume automated agents can indeed trigger rate limits or bans if it violates the usage policy. Be careful with request volume. Consider using a dedicated API tier or OpenRouter for safer, compliant access.

[miketsl]

Guys -- This is a pretty simple thing to understand. OpenClaw uses Pi-Agent as its agent harness. If you take your max plan, authenticate, and use this authenticated session in ANOTHER agent harness (not claude code) like pi-agent or opencode, you will get banned. If you use this authentication in an anthropic tool / harness like claude code there is no problem. So - if you build a system like nanoclaw or anything else, that essentially automates things by using claude code or agent sdk ... that is totally fine and good. What is not ok is using the authenticated session to drive OTHER people's harnesses like pi-agent or opencode or whatever.