← sources

Stop button didn't stop a Claude agent for 9 minutes

source: reddit (reddit:ClaudeAI)published 2026-11-18 01:27ingested 2026-11-19 02:36

retained text

Context:  
I'm on the Ultra plan, using Claude projects on macOS. Tonight a project thread (Fable 5.2, High effort) wouldn't stop.

What happened\*\*

\- At 6:40 PM I told it to stop. Over the next few minutes I pressed the stop button over and over and sent more than 10 STOP messages.

\- I resolved (closed) the thread. It kept running.

\- I quit the app on my laptop. It kept running, because it runs in the cloud.

\- My coordinator agent tried twice to stop it and according to it tried to forcefully stop it but was refused.

\- It finally stopped at 6:49 PM, about 9 minutes later, when it finished its step on its own.

\- it ate up a couple million tokens as well, it was really scary because in this case it was okay since it was just over not too important things but what if it were on something more important? Maybe a github repo or something that uses a very large amount of tokens?

\- Every time I manually pressed the stop button, it showed stopping... then would go right back to thinking



Isn't being able to control an ai agent (stop it when needed) the absolute bare minimum safety guardrails for AI?  


Questions:

1. Has anyone else seen the stop button fail to interrupt a running thread?   
2. Has as anyone gotten usage credited back for something like this?   
(I just want it back out of spite o.o)  
3. Has this happened to someone else?



document fetched 2026-11-19 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1xz2d3y/stop_button_didnt_stop_a_claude_agent_for_9