Automation Run Test fails on Grok 4.7: Failed to start background composer [not_found]

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Run Test on a scheduled automation fails immediately with Failed to start background composer: [not_found] Error when the model is Grok 4.7. The same automation, same environment, starts when the model is Grok 4.6. The only change between the failing and working runs is the model.

Steps to Reproduce

Steps to Reproduce

Open an existing automation on cursor.com/automations. Trigger: custom schedule. Environment: environment is unchanged
Set the model to Grok 4.7.
Click Run Test.
The dialog shows Failed to start background composer: [not_found] Error.
Set the model to Grok 4.6 and click Run Test again. The run starts.

Expected Behavior

Run Test starts the background composer on Grok 4.7 the same way it does on Grok 4.6.

Screenshots / Screen Recordings

Screenshot 2026-09-22 at 12.42.54.png

Operating System

Other

Version Information

Cursor Cloud, Web, no local clients involved.

For AI issues: which model did you use?

Grok 4.6 (working) and Grok 4.7 (not working)

Settings: non fast and xhigh for both, for 4.7 we choose 500k context size.

For AI issues: add Request ID with privacy disabled

None. The run never starts, so there is no bc-id. But see screenshot.

Additional Information

Automation in question: https://cursor.com/automations/f47c065b-b292-11f1-a3d8-362438fd9788

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Hey, thanks for the detailed report and screenshots. They make it clear that starting on Grok 4.7 fails with [not_found], while the same automation on Grok 4.6 runs fine.

This isn’t related to your automation or environment. Since this morning (UTC), Cloud Agent starts on Grok 4.7 have been getting rejected with [not_found], while other models are starting normally. This affects both Run Test and scheduled runs, so any schedule pinned to Grok 4.7 won’t run until the issue is fixed. Grok 4.7 in the editor itself isn’t affected. We’re tracking this.

Workaround for now: switch the automation model to Grok 4.6 or Auto, and runs should work again. Once things are back to normal, you can switch back to Grok 4.7 without any other changes.

I’ll post here as soon as there’s an update.