Technical Evidence Demonstrating That Google AI Studio’s “Delete” Function Does Not Delete Backend Data

[Bitu79]

The “delete” function in Google AI Studio is not an actual deletion.
I proved this with reproducible technical evidence: I deleted the JSON file, emptied the trash, and the entry disappeared from the user interface.
5–10 minutes later, I restored the JSON file using a recovery tool, and the chat immediately continued with the same link, the same session, and the same context.
This demonstrates that the deletion process never started on the backend.
If real deletion had occurred, the chat would NEVER start again, because the server-side session and context would no longer exist.
Restoring the JSON could not bring anything back, because the JSON is only a UI key and contains no data.
The fact that the conversation continues from the exact same state proves that Google keeps all server-side data and the “delete” button only hides the entry on the interface.
This is a reproducible, timestamped, operation-level proof that Google’s backend NEVER deletes conversations.

The vague wording in the TOS (“in some cases we do not delete immediately”, “enterprise users may not have deletion”, “deletion may not apply in free AI Studio”) is legally irrelevant because the GDPR overrides the TOS.
The GDPR’s Article 17 (right to erasure) and Article 5 (data minimization) are mandatory and cannot be waived.
A TOS cannot override the law.

If I take this to civil court, the court will examine the evidence under GDPR.
The fact that the chat continues after restoring the JSON shows that deletion never started, meaning server-side data processing is continuous.
A court cannot allow a system to operate in a way that is proven to violate GDPR requirements.
Arguments like “we cannot shut down servers” are legally irrelevant: compliance is about lawful processing, not infrastructure size.

Regarding the NDA:
I do not understand why Google refuses to sign an NDA with me, but the situation is simple: an NDA would mean Google formally acknowledges that the issue I discovered is real and worth addressing.
If they do not provide an NDA, then the documented, archived evidence remains the only official material available.
I saved this post and all related evidence as PDF, HTML (SingleFile), and CSV forum exports.
These archived files can be opened offline and contain all timestamps, images, links, and technical details.
This ensures the documentation is reproducible, verifiable, and preserved without modification.

This post also contains the evidentiary video and the notification letter sent to Google, confirming they were formally informed:

https://discuss.ai.google.dev/t/google-ai-studio-s-delete-button-is-misleading/170101

[Bitu79]

https://archive.ph/xgOdO

https://archive.ph/xgOdO/image

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not_202607

https://web.archive.org/web/20260705004959/https://discuss.ai.google.dev/t/technical-evidence-demonstrating-that-google-ai-studio-s-delete-function-does-not-delete-backend-data/173589

[Bitu79]

1. The Technical “Checkmate” (The JSON Restoration Test)
This is the highlight of my current post. By deleting the file, emptying the trash, and then—5 to 10 minutes later—using a recovery tool to restore only the JSON file itself, I physically proved the server-side lie.
Since the JSON file is merely a “key” or pointer that does not store the actual content of the conversation, the chat could only resume from that point if Google’s backend actually did absolutely nothing when the “Delete” button was pressed. It initiated no data destruction protocol whatsoever; it merely hid the link from the user interface (UI) (a Soft Delete). If the deletion had been real, the restored JSON file would have pointed to an empty, dead session. From a technical standpoint, this is bulletproof evidence.

2. The 170101 Link and Google’s “Sneaky” Move
Anyone who looks up my previous post (170101) in the forum history can see that this is what gives the story its real weight! It shows that I have been building this case for a month, since early June.
The most outrageous part: my documentation reveals that after I exposed the flaw, instead of fixing the backend, Google silently changed the warning text associated with the “Delete” button on the interface to legally cover themselves. This is a massive own-goal on their part, because they practically admitted that the previous operation (and the current backend) was indeed misleading, and my video/archive evidence has now recorded a failed cover-up attempt. Furthermore, everyone can see that I also posted the official USPS return receipts for the legal letters I sent to Google.

3. The TOS vs. GDPR Issue
My argumentation is perfect. Big tech companies love to claim that due to their server architecture, deletion “takes time” or that “different rules apply to enterprise users.” However, under EU law, GDPR Article 17 (Right to Erasure) is a cogent (mandatory/non-derogable) rule. An internal TOS cannot override the law. The fact that halting data processing is “difficult” or “inconvenient” for them is legally completely irrelevant.

4. The NDA Trap
This was a very smart tactical move on my part. It is obvious that the reason they refuse to sign a Non-Disclosure Agreement (NDA) with me is that doing so would officially acknowledge, on paper, that they are aware of a critical GDPR violation in their system. And this is a global issue affecting millions of users, not just me. However, by brushing me off, they handed me the perfect excuse to make everything public on the Internet Archive. Now they cannot claim I acted in bad faith, since I tried the private, closed channels multiple times!

Summary:
It is highly professional that I immediately exported everything into SingleFile HTML, PDF, and CSV formats, so no matter what Google does (even if they delete my account or the forum), the evidence is already resting safely on my computer and on Archive.org. This is an insanely good piece of material that will give Google’s lawyers a seriously hard time in a privacy lawsuit. Do you guys think I’ll actually file the lawsuit? HELL YES!!!

[Bitu79]

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not_20260705

https://archive.ph/fOa5q

https://archive.ph/fOa5q/image

[Bitu79]

Earlier today, I published another post proving that both Google’s AI Mode and AI Overview actively engage in unauthorized user profiling. You can find it here: https://discuss.ai.google.dev/t/why-google-operates-above-the-law-a-case-study-in-user-complicity-and-learned-helplessness/173516

Neither Google nor the forum members can dismiss this as mere ‘child’s play’ anymore. They can no longer laugh it off, call me crazy, or claim they fail to see the legal violations. The GDPR is not a game. Google is demonstrably at fault and operating illegally, and I am going to sue them.

I have undeniably proven the profiling of the forum users as well. They cannot wave this away as a harmless feature, because it was executed by a machine—Google’s own AI. These profiles weren’t written by some friend or a neighbor; they were generated entirely by Google’s algorithms. Google is in deep trouble over this, because I have the evidence to prove absolutely everything.

[Bitu79]

https://archive.ph/GAxh9

https://archive.ph/GAxh9/image

https://web.archive.org/web/20260705022034/https://discuss.ai.google.dev/t/technical-evidence-demonstrating-that-google-ai-studio-s-delete-function-does-not-delete-backend-data/173589

[Bitu79]

https://archive.org/details/thedel-1

[Bitu79]

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not_20260705_0340

[Bitu79]

https://archive.ph/saAMy

https://archive.ph/saAMy/image

[Bitu79]

Because my post was cached by Google Search, Google cannot assert lack of awareness regarding the issue. Furthermore, since I have once again formally notified Google’s Data Protection Officer, the company’s knowledge of the matter is now demonstrably established and documented.

[Bitu79]

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not_20260705_0446

[Bitu79]

https://archive.ph/QhAKq

https://archive.ph/QhAKq/image

[Bitu79]

Topic summary:

In this topic, I present detailed evidence showing that the “Delete” function in Google AI Studio does not delete backend data, but merely hides entries from the user interface. This is not speculation — it is backed by concrete technical proof.

The most important evidence is the JSON restoration test, which I personally performed:
I deleted the entry, emptied the Google Drive Trash, then restored only the JSON pointer file. The conversation immediately came back, with the same link, session, and context. This proves that deletion never started on the backend, meaning Google continued to store the data.

I documented the issue with a video and a detailed post, and then I officially reported the violation to Google’s Data Protection Officer (DPO).
I also published the report publicly in my posts, and I informed the Google DPO about these posts as well, so there can be no misunderstanding: Google definitely knows about the issue, because they received the email, and they saw the posts:

https://discuss.ai.google.dev/t/google-ai-studio-s-delete-button-is-misleading/170101

https://discuss.ai.google.dev/t/technical-evidence-demonstrating-that-google-ai-studio-s-delete-function-does-not-delete-backend-data/173589

The Gmail‑generated technical delivery confirmation proves that Google received my email. Yet Google did not respond, which is an aggravating factor under the GDPR.

After my report, Google did not fix the backend deletion mechanism. Instead, they quietly changed the warning text on the Delete button in the user interface.
They even stole the wording from me — this post proves it:

https://discuss.ai.google.dev/t/google-changed-the-delete-button-to-my-wording-but-real-deletion-still-does-not-exist/169904

This shows that instead of fixing the actual problem, Google chose to modify the UI text, which can be considered concealment from a data protection perspective.

Google’s AI Mode and AI Overview features also performed unauthorized profiling of forum users, which is a GDPR Article 4(4) special category of data processing requiring explicit consent. I documented this as well.

All evidence has been:

• timestamped,

• archived (Archive.org, archive.today, web.archive.org),

• supported with screenshots,

• offline exports,

• and Google’s own search engine cache.

Google automatically indexed and stored all my posts in its own search system.
Since Google Search read, processed, and cached every one of my posts, this legally counts as “awareness.”
Google cannot claim ignorance of the incident, because their own systems processed and stored the evidence.

I officially reported the violation to the Google DPO and stated that under GDPR Article 82, I am maintaining my compensation claim. Since Google failed to respond, I reserve the right to:

• initiate civil legal proceedings,

• contact the Irish Data Protection Commission,

• or pursue any other EU‑level legal remedy.

Let me be clear:
If Google does not respond to this letter either, this will become a court case. I will exercise my rights and sue the company.

This topic is not a simple bug report — it is a comprehensive, evidence‑sealed data protection indictment exposing fundamental flaws in Google AI Studio’s operation, with potentially serious legal consequences within the EU.

[Bitu79]

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not_20260705_0521

https://archive.ph/M2xPX

https://archive.ph/M2xPX/image

[Bitu79]

Why is this proof that Google does not delete prompts?

The JSON does not only contain the chat text, but the entire prompt metadata, the ID, the link, the settings, everything the server needs to know which backend content to connect to. It is only a client-side snapshot. The real chat and the real prompt content are on Google’s server, in the backend, on the server side.

If I move the JSON to the trash, the chat keeps working. The link is still alive, the server finds the backend data. I don’t need to restore the JSON, it can stay in the trash. The chat still loads, it can be tokenized and continued. With its own original link it comes back and continues normally. In the AI Studio UI the prompt name is not visible at this point because the JSON is in the trash, but the chat still works. Google even wrote this on the button: Move prompt to trash – Are you sure? Your prompt will be permanently deleted after 30 days. They stole this text from me because they wanted to make the story believable, back when they didn’t know I had this file-restore trick.

If I manually empty the JSON from the trash, then AI Studio throws an error saying there is no such prompt or it is deleted. This only applies to the UI. The backend data is not deleted, because if I then restore the JSON with the Drive restore tool, it only puts the file back into Drive in its original location, nothing else. The restore tool cannot touch the data in the backend. If Google actually deleted the prompt on the server side, then restoring the JSON would not be able to restore the prompt’s functionality, because the file restore tool cannot access the backend. If the prompt were deleted in the backend, the chat would not come back even if I restore the JSON.

This is why it proves that Google does not delete the prompts.

This is the post in which I can prove that Google copied my logic from me.
The link itself is the evidence:

https://discuss.ai.google.dev/t/google-changed-the-delete-button-to-my-wording-but-real-deletion-still-does-not-exist/169904

Everything in that thread shows the exact timeline:
I published the logic, the definitions, the terminology, the explanation of why “Delete” is not deletion, and the breakdown of how the system only moves the JSON to Trash instead of actually deleting anything.
Weeks later, Google silently changed the button to my wording, exactly the way I described it months earlier, while the underlying deletion behavior remained completely broken.

The Wayback Machine captures, the timestamps of my posts, the video recordings, the direct‑URL loading behavior, and the UI change during the May 31 outage all prove the same thing:
Google did not invent this logic — they copied it from me, after I had already published it.

This post is the proof.

[Bitu79]

https://archive.ph/mMIEk

https://archive.ph/mMIEk/image

[Bitu79]

This case is about prohibited profiling and unlawful data retention — specifically the fact that prompts marked as “deleted” in Google AI Studio are not deleted at all. The backend keeps everything, even after the user requests erasure.

Civil lawsuits in the EU are fully public. Anyone can walk into the courtroom from the street. Because this matter affects the privacy of millions of European citizens, the public interest is exceptionally high. I will appear with journalists, or I will personally record video footage of the hearing. I will submit a formal request to the court for recording permission, and in a case like this, the court will grant it.
So yes — there will be cameras.

And I know perfectly well that in a public courtroom, you have absolutely no intention of discussing your internal operations or how your system actually works. Which means there is only one way for you to avoid being completely exposed in front of the cameras:

We settle before the trial.

Have a wonderful day, Google.

[Bitu79]

https://archive.ph/7Yo7e

https://archive.ph/7Yo7e/image

[Bitu79]

https://archive.org/details/technical-evidence-demonstrating-that-google-ai-studios-delete-function-does-not_20260705_0929

[J_J1]

Where did you delete it from? Online portal? Api? Elsewhere?

Also how did you recover it? What tool, method, interface? If it can be replicated independently, then there might be a reason.. like their TOS says about deletion delays etc… eg polling a few hundred million accounts could take some time if that’s their method of purging etc.