Cursor Agent Requests Randomly Blocked by Model Provider

Support Help
1

cursorbug
cursorbug446×628 36.9 KB
For the last 3 days, Cursor is frequently blocking my Agent requests with a “Request blocked – model provider’s usage guidelines” message. I’m working on a legitimate backend project and only implementing checks/features according to my client’s requirements. The requests are normal development tasks, but they are being randomly flagged even when there is no harmful, abusive, or unauthorized intent. This is disrupting development and happens repeatedly across different requests.

5

Hey, thanks for the report and for the screenshot with the Request ID.

Your prompt looks fine. This isn’t something in your project or setup. The block is coming from a content filter on the model side that Auto picked for these requests. Over the last few days, that filter has been mistakenly flagging some normal dev prompts like pasted logs, stack traces, hostnames, and DB status or field logic as sensitive. This is a known issue, we’re tracking it with the provider, but I can’t share an ETA yet.

What should unblock you right now:

  1. In the model picker at the bottom of the chat, switch from Auto to a specific model for these requests, like Claude or GPT. You can switch back to Auto later.
  2. If you want to stay on Auto, trim pasted logs and error output down to one or two key lines and remove hostnames and ports before sending.

If the issue happens again on a specific model already, not through Auto, send a fresh Request ID using right upper corner of chat > Copy Request ID in the error block and we’ll take a closer look.

7

this is happening a lot with the cursor grok models for prompts that are very much not sensitive. Please fix ASAP

8

I’m am having this same issue constantly with Grok 4.7 it is driving me nuts!

9

Hey, thanks for reaching out. What you’re seeing is the same model-side content filter that’s been incorrectly flagging some normal dev prompts over the last few days like pasted logs, stack traces, hostnames, and DB status or field logic as sensitive. It’s not anything in your project or setup. We’re aware of the issue and tracking it with the provider, but I can’t share an ETA yet.

What should unblock you right now:

  1. In the model picker at the bottom of the chat, switch from Auto or Grok to a specific non-Grok model like Claude or GPT for these requests. You can switch back later.
  2. If you want to stay on the current model, trim the pasted logs and error output down to one or two key lines, and remove hostnames and ports before sending.

If this happens again on a specific non-Grok model not via Auto, send a fresh Request ID from the error block using the top-right chat menu > Copy Request ID, and we’ll take a closer look.