Hello,
We are experiencing a production-blocking issue with an application created and deployed through Google AI Studio.
The application was working correctly in production, including server-side Firebase Admin SDK access to a named Firestore database.
On October 15, 2026, we have confirmed a successful production execution of our booking endpoint. The server-side Firebase Admin SDK successfully executed a Firestore transaction read (tx.get(...)) and completed the request successfully.
The same deployed Cloud Run revision is still serving production traffic, but the same Firebase Admin Firestore operation now fails with:
PERMISSION_DENIED
gRPC code: 7
The failure occurs on the first Admin Firestore transaction read.
We compared the last known working state with the current failing state. The relevant application-side configuration appears unchanged:
-
Same Cloud Run service and revision
-
Same Firebase Admin initialization
-
Same named Firestore database
-
Same Application Default Credentials mechanism
-
Same runtime identity configuration
-
Same Firebase Admin version
-
Same transaction implementation
-
Same relevant application configuration
We have not changed Firestore Security Rules as a workaround because this is a server-side Firebase Admin SDK operation.
A second problem is now preventing diagnosis/deployment
We added a minimal one-shot startup diagnostic that performs a single Firebase Admin Firestore read with zero writes.
However, we cannot deploy it because AI Studio Republish now fails before a new Cloud Run revision is created.
Using the normal AI Studio Republish workflow, the UI reports:
Failed to generate API key, The service account creation failed. Please try again.
After this failure:
-
No new Cloud Run revision is created.
-
The existing production revision remains active.
-
The diagnostic code therefore never reaches the runtime.
-
Repeating the normal Publish action does not produce a new Cloud Run revision.
The project is an AI Studio managed/Starter Tier project. Our user account does not have full project-level administrative permissions, so manually changing the underlying IAM or managed service-account configuration is not an appropriate or available workaround.
At this point we appear to have two potentially related managed-platform issues:
-
Firebase Admin Firestore access that previously worked from the existing production runtime now returns
PERMISSION_DENIED. -
AI Studio can no longer Republish the application because service-account/API-key provisioning fails.
We are NOT claiming these two failures definitely have the same root cause, but the timing and managed-identity involvement make this worth investigating.
Could someone from the Google AI Studio team please advise whether there is a known issue with managed project identity/provisioning, or help us determine how to restore the managed deployment state?
We can provide project-specific identifiers privately to Google staff if required, but have intentionally omitted project IDs, service-account addresses, API keys, and other sensitive information from this public post.
This is currently blocking our production booking functionality and also preventing us from deploying a diagnostic or corrective revision.
Thank you.
