Safe prompt blocked cause of the model provider's usage guidelines

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I get this message in a lot of my requests and I don’t think I am using a sensitive prompt. Here is the error/notification :

Request ID: 3d94c9ab-cfa5-4700-a42c-301c2c828a11
{“error”:“ERROR_OPENAI”,“details”:{“title”:“Request blocked”,“detail”:“We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:,“analyticsMetadata”:{“actionRequired”:“cyber_policy_review”}},“isExpected”:true}
Request blocked We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.
ActionRequiredError: Request blocked We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.
at f6f (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8346:217030)
at _to (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8346:215869)
at U6f (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8347:13639)
at Object.classify (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8347:17861)
at P6f (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8347:5924)
at async Dto.run (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8347:17168)
at async Sle.runAgentLoop (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8447:30991)
at async FFo.streamFromAgentBackend (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:17887:16374)
at async FFo.getAgentStreamResponse (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:17887:37742)
at async bDe._submitChatMaybeAbortCurrent (vscode-file://vscode-app/c:/Users/AggelosDrampas/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8425:4549)

Steps to Reproduce

Just this prompt to my agent:
I get this error in the opik-backend:

ERROR: Exception Details
ERROR: Exception Primary Class: Throwable
ERROR: Exception Primary Reason: std::exception. Code: 1001, type: std::exception, e.what() = std::exception (version 26.3.16.16 (official build))

ERROR: Exception Primary Source: ClickHouse 26.3.16.16
INFO [2026-09-16 08:39:32,741] liquibase.command.CommandScope: Logging exception.
INFO [2026-09-16 08:39:32,742] liquibase.command.CommandScope: Command execution complete
Running database migrations…
Current directory is: /opt/opik
OPIK_VERSION=2.2.53
liquibase.exception.CommandExecutionException: liquibase.exception.DatabaseException: liquibase.exception.DatabaseException: ClickHouse exception, code: 1002, host: opik-clickhouse, port: 8123; std::exception. Code: 1001, type: std::exception, e.what() = std::exception (version 26.3.16.16 (official build))
[Failed SQL: (1002) INSERT INTO default.DATABASECHANGELOGLOCK (ID, LOCKED) VALUES (1, 0)]
at liquibase.command.CommandScope.lambda$execute$6(CommandScope.java:310)
at liquibase.Scope.child(Scope.java:225)
at liquibase.Scope.child(Scope.java:201)
at liquibase.command.CommandScope.execute(CommandScope.java:251)

Expected Behavior

The agent responds.

Operating System

Windows 10/11

Version Information

Version: 3.20.21 (user setup)
VS Code Extension API: 1.128.0
Commit: f09fca384ceca23f7bf21f9c23655b162641d740
Date: 2026-09-13T19:02:51.781Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 42.10.0
Chromium: 148.0.7778.280
Node.js: 24.18.1
V8: 14.8.178.38-electron.0
xterm.js: 6.1.0-beta.291
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Cursor Grok 4.6 , Medium

For AI issues: add Request ID with privacy disabled

Request ID: 3d94c9ab-cfa5-4700-a42c-301c2c828a11

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Same, getting this error message

We are unable to complete this request because it was blocked under the model provider's usage guidelines. Try a less sensitive prompt.

Here was my prompt:

why are these hand created? That is an antipattern entirely of our goal. the backend is supposed to generate the API paths and data types. Explain the delta.

```

Request ID: 19825b68-d17f-4520-9c80-9bfb59d2bd0c{"error":"ERROR_OPENAI","details":{"title":"Request blocked","detail":"We are unable to complete this request because it was blocked under the model provider's usage guidelines. Try a less sensitive prompt.","isRetryable":false,"additionalInfo":{},"buttons":,"planChoices":,"analyticsMetadata":{"actionRequired":"cyber_policy_review"}},"isExpected":true}Request blocked We are unable to complete this request because it was blocked under the model provider's usage guidelines. Try a less sensitive prompt.ActionRequiredError: Request blocked We are unable to complete this request because it was blocked under the model provider's usage guidelines. Try a less sensitive prompt.at hdf (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9066:9274)at Wqi (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9066:8425)at wcp (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9067:11962)at Object.classify (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9067:16467)at gdf (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9067:5856)at async fWo.runWithSharedTurnRunner (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9067:15819)at async xz.runAgentLoop (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:20459:3067)at async lba.streamFromAgentBackend (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:20487:14009)at async lba.getAgentStreamResponse (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:20487:23432)at async VIe.submitChatMaybeAbortCurrent (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:9229:2399)at async wk (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:19518:83910)
6

Hey, thanks for the report and for including the Request ID.

The prompt is really harmless. This isn’t something in your setup, and you didn’t do anything wrong. The block is coming from the model provider’s own content filter for Grok 4.6. Right now it’s flagging some normal prompts, including pasted logs, stack traces, hostnames, and ports, as potentially sensitive. We’re tracking it, but I can’t share an ETA yet.

Until it’s fixed, there are two workarounds:

  1. In the model picker, switch to a non-Grok model for these prompts, like Claude or GPT, then switch back.
  2. If you want to stay on Grok 4.6, trim the pasted log down to one or two lines with the actual error, here that’s ClickHouse exception, code: 1002 and the failed SQL, and remove hostnames and ports.

@cfix, you have the same symptom. Switching models should help in your case too.

If the issue still happens after an update, send a fresh Request ID and we’ll take another look.

8

Same here. Request ID: 210e9d65-6a4e-4754-8435-cd9a5dd6d725

I tried to tell the model i’m signing off on an idea by saying “alright lets do it” and it interpreted that as sensitive…

11

Same here:
Request ID: 3da157cc-abae-4b50-9087-ffb6d3c55a79
{“error”:“ERROR_OPENAI”,“details”:{“title”:“Request blocked”,“detail”:“We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:,“analyticsMetadata”:{“actionRequired”:“cyber_policy_review”}},“isExpected”:true}
Request blocked We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.
ActionRequiredError: Request blocked We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.
at NPf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6924:18135)
at ZVs (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6924:16789)
at lLf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6925:13912)
at Object.classify (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6925:17472)
at iLf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6925:6366)
at async aqs.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6925:16779)
at async Ule.runAgentLoop (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8318:31468)
at async $Uo.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:17780:17013)
at async $Uo.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:17780:39121)
at async ADe._submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8296:4491)

12

Hey @Erez_Makavy, thanks for the report and for the Request ID.

This is the same symptom as above in the thread, and it’s not related to your prompt or setup. The block is coming from the model’s own content-filter provider, which is currently flagging some normal requests (logs, stack traces, hostnames, ports, and sometimes even plain text) as potentially sensitive. We’re tracking it, but I can’t share an ETA yet.

Until it’s fixed, there are two workarounds:

  1. In the model picker, switch to a non-Grok model (for example Claude or GPT) for those requests, then switch back.
  2. If you want to stay on Grok, trim the paste down to one or two lines with the key details, and remove extra context like hostnames and ports.

If the issue happens again after an update, send a fresh Request ID and we’ll take another look.

13

same thing - and that was too often at last on Grok 4.7 or 4.6 model.
“We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.”

Request ID: ba52f98a-9cfb-42c6-ac4b-243c6e9626a1
{“error”:“ERROR_OPENAI”,“details”:{“title”:“Request blocked”,“detail”:“We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:,“analyticsMetadata”:{“actionRequired”:“cyber_policy_review”}},“isExpected”:true}
Request blocked We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.
ActionRequiredError: Request blocked We are unable to complete this request because it was blocked under the model provider’s usage guidelines. Try a less sensitive prompt.
at e6f (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8151:216870)
at Zio (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8151:215709)
at T6f (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8152:14075)
at Object.classify (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8152:18296)
at _6f (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8152:6340)
at async cro.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8152:17603)
at async kle.runAgentLoop (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8252:31373)
at async J6o.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:17701:16852)
at async J6o.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:17701:38909)
at async lDe._submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:8230:4456)

14

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

This is the same symptom that’s already being discussed earlier in the thread, and it’s not related to your prompt or setup. The block is coming from the model’s own content filter provider, which is currently flagging some normal requests (logs, stack traces, hostnames, ports, sometimes even plain text) as potentially sensitive. We’re tracking it, but I can’t share an ETA yet.

Until it’s fixed, here are two workarounds:

  1. In the model picker, switch to a non-Grok model (for example Claude or GPT) for those requests, then switch back.
  2. If you want to stay on Grok, trim the paste down to 1 to 2 lines with the key info, and remove extra context like hostnames and ports.

If the issue happens again after an update, send a fresh Request ID and we’ll take another look.

15

this one “trim the paste down to 1–2 lines with the key info” doesn’t work.
one way to fix this is to switch from Agent to Composer 2.5. (I’ve already used the other limits).

16

sorry for spamming but I want to point out that it happens more often when I switch from Ask mode to Agent mode. mb it will helpful f u. grac!

17

I noticed that too.. It’s all Grok’s doing I believe, when I switch model it starts working just fine. Ever since they got acquired they are getting worse and worse and everything is pointing at Grok.

Now when you set to Auto sometimes it no longer asks question in a separate questions panel with options, instead it just responds with question, and then we have to type and answer the questions rather than selecting from options. As soon as you switch to some model such as Composer it starts asking questions just fine.

Waited over a month for this Git hooks issue.

Been using Cursor for over a year, never had any complaints until last few months.

The downgrade of Cursor has started, if this keeps going we (our team) might think of switching to some other provider.

19

Hey, thanks for the extra details and the Request ID, it helps.

Confirmed: the most reliable workaround is to switch in the model picker to a non-Grok model. Composer 2.5, Claude, or GPT usually work fine for these requests, then you can switch back. @deniro, thanks for the callout, if trimming the log doesn’t help, switching the model is still a working option.

I’ve passed along your note that it triggers more often when switching Ask → Agent, along with the rest. The block itself is on the model provider’s content filter side, it’s not related to your prompts or your setup. We’re tracking it, but I can’t share an ETA yet.

@Shashank_Salian, the Auto mode behavior with the questions panel, where questions show inline instead of in a separate panel with options, is a separate issue and not related to this block. Please start a new thread for it with a Request ID and a short set of steps to reproduce, otherwise it’ll get lost here among the block reports.

If the block keeps happening after updating, please share a fresh Request ID and we’ll take another look.