I am submitting a formal complaint regarding the deletion of one of the key files of my messenger project while I was using your service.
I requested a limited modification to my project, but instead a critical file was removed. This file was essential to the functionality and development of the application, and its deletion caused serious disruption, loss of time, and damage to my work.
At this stage, I am requesting a full investigation into what happened. Based on the circumstances, it appears that the file may have been deleted as a result of an erroneous action during the modification process. However, I need your team to verify this through your internal logs and records.
I request the following:
A full investigation into the incident.
Clarification of how and why the file was deleted.
Review of any available logs, command history, or execution records related to the session.
Information on whether the deleted file can be recovered from backups, temporary storage, snapshots, or other recovery sources.
A clear explanation of what safeguards you have in place to prevent such incidents from happening again.
This was one of the core files of my messenger project, and its deletion has had a serious impact on my development process.
Please treat this matter as urgent and provide a substantive response as soon as possible.
I am writing to express my serious concern and request a refund.
This is now the fourth time I have encountered the same problem: your model accesses files that it was not asked to open and, through shell actions, deletes files. This is unacceptable behavior for a paid product.
I have already requested a proper refund because this is not an isolated incident. Yes, I have backups, but I have never experienced problems like this with any other product I have paid for. I should not have to spend my time worrying about how to protect my project from your unstable model instead of focusing on my actual work.
A tool like this is supposed to help with development, not create a constant risk of file loss or project damage. The fact that this has now happened four times has completely undermined my trust in your service.
I am requesting:
a full refund;
a review of these repeated incidents on my account;
an explanation of why the model is accessing files it was not instructed to open;
clear safeguards to prevent unauthorized file access and file deletion in the future.
Please treat this matter seriously and respond promptly.
I am contacting my bank and will be disputing the payment transaction, because your model creates an unacceptably high risk to my work.
Even leaving aside the file deletion issue, your model does not properly account for text encoding. As a result, instead of normal text, I often end up with corrupted output that I have to manually fix every time. The model uses PowerShell Set-Content to write code into files, and on Windows these scripts often save text using the default system encoding, such as ANSI/Windows-1251. After that, when I ask the model to fix the encoding problem, it creates a Python script that causes even more damage and breaks the code completely.
This is only one example. As an analyst, I can clearly see that the model is not suitable for production use in its current state, because its actions themselves require constant testing and verification.
There is also the issue of billing. My subscription included 25,000 credits, yet when your servers were down, every attempt to continue the conversation still consumed credits. In other words, I was being charged even when the service itself was not functioning properly.
For these reasons, I consider the service unreliable and unsafe for my work, and I am proceeding with a dispute of the payment transaction.
Bro don’t get me wrong I’m quite upset with Google’s recent handling of their quota and such. But why did you approve the model to delete a file? You have to manually click approve on all the changes by default. The only way the model can manually delete a file without your permission is if you approved the rm (or pwoershell remove-item) command, or if you hit approve on a deleted file, then you did git add . Then you did git commit THEN you did git push.
Again not trying to be a Google shill but did you change all the settings to auto proceed? Don’t do that…
No other AI agents I have used behave like this. I do not see this kind of problem anywhere else. What makes it worse is that your team did not even properly respond to the complaints about the product, which is simply unacceptable treatment of users.
And regarding your point that I gave permission: of course I gave permission for the work to be performed. That is logical. But when the model starts making changes and then says, “I accidentally damaged your project while applying the fix,” that is a completely different matter. I have never encountered anything like this before.
It is obvious that I would never intentionally approve a destructive command such as rm, and my project was not on GitHub — it was stored locally on my laptop.