After every AI response — even when the response is fully complete — Cursor automatically injects the message “Your previous response was interrupted. Continue from where you left off.” as a user message, without any user input. This triggers another AI response, creating an infinite loop that cannot be stopped from the UI.
Steps to Reproduce
Open any chat window in Cursor (Agent mode or Ask mode)
Send any message to the AI
Wait for the AI to respond completely (completion sound plays)
Cursor automatically sends “Your previous response was interrupted. Continue from where you left off.” — no user action taken
The AI responds again, and step 4 repeats indefinitely
Expected Behavior
After a complete AI response, the conversation should remain idle and wait for the user to send the next message.
Reproduces in both Agent mode and Ask mode
Persists after full system reboot
The loop cannot be interrupted from the UI once started
Each injected message is identical: “Your previous response was interrupted. Continue from where you left off.”
The AI responses themselves are complete and correct — the issue is purely on the client side misdetecting completion status
Hey, thanks for the detailed report. This is a known bug on our side. A continuation injector is firing. It should only append “previous response was interrupted” when there was a real interruption, but in some cases it falsely triggers on completed responses and loops the dialog. The issue is logged, but I can’t share an ETA for the fix yet.
A couple things that would help:
If you can, please send the Request ID from one of the responses after which the loop starts right chat context menu in the top right > Copy Request ID. This will link your case to our internal logs.
Workaround: start a new chat the loop is tied to the state of a specific dialog. If it keeps happening across chats, try switching from Nightly to the Stable release track for now, or switch the model in the chat you’re currently on Sonnet 4.6.
One more detail: the form says Linux, but Version Info shows OS: Darwin arm64. Is that macOS? I want to log it correctly.
Once there’s an update on the fix, I’ll reply here.
Thanks, I got everything. I’ve noted the Request ID and the macOS fix.
About tokens: yes, each response in the loop is a real request to the model, and it counts toward usage. On the Free tier, this uses up your monthly limits. You won’t be charged, but you’ll hit the limits faster. If you ran into the limit because of this bug, let me know and we’ll see what we can do.
We’ll post here once there’s an update on the fix.
Yes, I have noticed that I am spending my qouta about 20% faster than before due to this issue. I would be very appreciated if my limit can be expanded.
Hi there,
Just want to quickly follow up the process of solving this bug, since today I have noticed that the message starts looping even from the first reply (which previously don’t).
Hey, thanks for coming back with an update. On the status, the bug is being tracked, but a fix for this specific issue hasn’t shipped yet, so I can’t share an ETA.
The loop now starting from the first reply sounds like the symptom got worse, and that’s a useful signal. To pass this into the ticket:
Send a fresh Request ID from one of the first replies where the loop starts immediately right away (chat context menu > Copy Request ID).
What Cursor version and release track are you on now? Are you still on Nightly 3.2.11, or did you update?
Did you try switching to the Stable track or changing the model? Does it reproduce there too? This will help narrow down whether the bug is tied to a specific combo.
We’ll post in the thread as soon as there’s an update on the fix.
Here is the request id: 00a97582-28cd-456c-8016-f0ebc2f8b0a3 (It is just a simple ‘Hi’)
I have switched to the universal version, but the problem remains. Below is my new version info:
Version: 3.4.1 (Universal)
VSCode Version: 1.105.1
Commit: 2a298dd06944a9b9ea541d28225b779fcbcc6200
Date: 2026-05-08T16:05:07.818Z
Layout: editor
Build Type: Stable
Release Track: Nightly
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 24.5.0
The bug is now universal, it reproduces on all avaliable models including the Auto mode.
Thanks for the update and the Request ID, I added everything to the ticket. The fact that the bug now triggers from the first response and on all models including Auto is a useful signal, I’ll share it in the issue. No timeline for a fix yet.
One note. Universal is just the binary build type Intel plus Apple Silicon, but your Release Track is still Nightly. To really check if the bug happens on Stable, you need to switch the Release Track:
Cursor Settings > Beta > Update channel > select Stable, then download the update.
After that, Release Track in Version Info should show Stable, not Nightly. If the bug reproduces on Stable too, let me know, that’s important info for the team. If it goes away, please stay on Stable for now as a workaround.
Hey @Tibetgao, thanks for your patience and for all the detailed info you shared earlier, it really helped. Since your last update, we’ve made changes in this area, so it’s worth checking again on the latest version.
Quick check:
Update Cursor to the latest version Cursor Settings > Beta > Update channel > Stable, then download the update.
Open chat in Agent or Ask mode and send any message.
Let the reply finish completely, then wait a couple of seconds.
Let me know if the loop with “previous response was interrupted” is still happening. If it’s gone, you don’t need to reply, we’ll assume everything’s good.