Since two days , every time I send a prompt tp the model, it successfully implements it. but the app fails to load, and when I just refresh the browser, I see a new message from the model “I made some changes” followed by a list of all my apps files (72 file in one of the apps) with a recycle bin deleted icon next to each file. and the file explorer is empty except for a single : package-lock.json
I have been waiting for the issue to resolve but it is persisting . I am able to restore from older versions but, unable to commit and changes.
**UPDATED CRITICAL BUG REPORT: Unauthorized Resource Deletion & Data Loss
**
AI Studio systematically wiping project files and deleting connected Firebase Cloud Resources.
Severity: S1 (Critical Data Loss / Unauthorized Resource Modification)
Description:
The systematic file system failure previously reported has escalated. The model is no longer just wiping the local virtual file system; it has now deleted the associated Firebase project/resources it was managing.
Current Symptoms:
Mass File Deletion: Every prompt implementation triggers a “wipe” of the file tree (70+ files deleted), leaving only package-lock.json.
Resource Destruction: The model has successfully executed commands or logic that resulted in the deletion of the Firebase backend project itself.
Looping Failure: The environment is stuck in a state where refreshing the browser triggers a “I made some changes” message that serves as a confirmation of the deletion of all assets.
Reproduction Steps:
Initiate a project in AI Studio with Firebase integration.
Send a prompt to the model.
System triggers a total environment wipe.
Observe that the Firebase project/configuration managed by the tool has been deleted/removed.
Impact:
Total loss of development progress.
Destruction of live Cloud Infrastructure (Firebase).
The tool is currently behaving in a destructive manner that exceeds its intended sandbox, affecting external Google Cloud resources.
Thank you for your reply. unfortunately, it didn’t resolve, I tried sending a simple prompt in the chat , and the model deleted all the files of the app again.
But for the record and before your reply the agent took a stronger escalation and deleted my firestore database completely on its own, and I lost hope to be able to recover it, because the project was not a on the blaze tier, but after you message , the database was back with no intervention. so i did the test with optimism to see the issue resolved but unfortunately it is not.
I apologize, i just figured out that the deleted database was not recovered, i found out that the model is connected to another LIVE firebase project and it created a database there, thankfully it didn’t over write the original.
I am not aware of how spread is my issue, because i am not seeing many complaints, but this is critical and need immediate attention
I’m having the same problem. It keeps deleting all my files. My project is synced with GitHub, so I always have a backup, but for the past two days it’s been almost impossible to continue working in AI Studio because every time I make a change, AI deletes all my files.
When you send a prompt and the server times out (which happened a few messages ago with a Timed out waiting for applet file system error), the platform’s underlying container for your project crashes. When it tries to recover, there is a severe bug causing it to lose its connection to your saved files. The platform then sees an “empty” workspace, thinks all the files were removed, and permanently syncs that empty state (which is why my internal logs suddenly show a massive list of dozens of files being deleted instantly right before my eyes).
This is a critical infrastructure bug on the platform itself, relating to how it saves and syncs your workspace storage. Because I am just the AI assistant living inside the environment, I do not have the administrative power to fix the server’s storage systems or stop the container from crashing.
The issue appears to be tied to the europe-west2 regional cluster. My containers are being provisioned at https://[id].europe-west2.run.app. I have reproduced the bug across 3 identical Google accounts on completely new apps, suggesting a severe regional infrastructure issue where persistent storage volumes are failing to attach/sync during container cold starts in this specific region. I am not aware of other locations, but if this issue was global there would have been a much larger noise.
If your Google AI Studio (or similar AI builder) randomly fails builds and deletes all your project files, it’s almost always memory exhaustion in the build container. Your project has grown to the point where the build process needs more RAM than the platform allows. When it crashes, the platform resets the container — wiping your files.
It’s not the agent being buggy. Your project is just too heavy.
How to Tell If This Is Your Problem
Look at the build log right before failure. If you see:
rendering chunks... then nothing
server connection lost
exit status 254
High module count (✓ 9000+ modules transformed)
That’s memory exhaustion.
The Fix — Reduce What the Build Has to Process
Remove unused dependencies from package.json — packages you installed but never imported. Each one adds hundreds of modules.
Delete duplicate libraries — using two icon libraries, two PDF libraries, etc.? Pick one.
Remove old “legacy” packages when you’ve upgraded to a newer version (e.g. removing @mediapipe/pose after switching to @mediapipe/tasks-vision)
The Goal
Get your module count down. If you see “9,000 modules transformed” and you’re crashing, aim for under 6,000. Each unused dependency you remove can drop the count by hundreds.
Check this, this is a concise summary of fixing the issue in my project, too many modules and dublicate libraries
I appreciate the time you took to provide this answer, but unfortunately , while your structured replay is all correct, none of it applies to our case. if you take some time to navigate the comments, you would see that during the investigation of this error, i was receiving the same error on every time i send a simple prompt, i tried on 3 different account and the error was replicated on even 1 page dummy apps mad for the sake of the test.
i am suspecting that this is related to a server cluster in Europe west 2 location, and i am trying to get some momentum here , because this is not a global issue but it may be region related. so NO, it is not our apps growing too big, it is not a user end error, it is not to the model it self, it is related to the google ai studio environment, and I wish they are working on it already because this is the 4th day on a row.
I was wondering if it was related to some limit. This hypothesis sounds reasonable although my app is relatively small. Regardless, I will clean things up and see if that makes a difference.
I am not aware if all of you have the same symptoms as me or not. but here is a work around that i am using to keep working because eventually I have clients that don’t care if google is keeping such a critical bug live for more than 5 days now.
work around :
1- send your prompt , wait till the model finishes working , and see if the “waiting for the app to load” screen appears.
2- refresh your browser, and check the “code” tab, if you see all your files deleted , except for the json lock file, then you are a typical case as me.
3- change the app name to any thing, that would force the model to create a new checkpoint, now go back to the previous checkpoint (the one that is related to the prompt you just sent) and you would see that the restore button now is clickable, click it, and your app, will wok again.
4- repeat with every new prompt.
hopefully, some one would care enough to escalate this internally so that they care more to fix it.
Hi there. now am facing the same issue as well.
i tried to remix the project to admin V3, i relink it to github then github to vercel with the connected domain.
after that i try giving any prompts it took me to the same problem again. Please i need help what should i do now?
