This screenshot was taken in Google AI Studio.
A completely new chat.
No history.
No previous messages.
Nothing carried over.
The only thing turned on was “Grounding with Google Search.”
And the moment I asked “Who is bitu79?”, the model immediately pulled data about me from the internet and generated a profile:
-
it connected my nickname to my real name
-
it pulled forum posts, Reddit activity, old accounts
-
it merged everything into one identity
-
and presented it as a structured description of “who I am”
This happened in a fresh chat, without any prior context.
It didn’t come from memory.
It didn’t come from previous prompts.
It came directly from Google Search, in real time, because Grounding was enabled.
This is profiling.
And the screenshot proves it.
The forum’s own summary gives Google another own goal:
POST — “Here’s the proof: the model does see my email address”
Here is the proof.
This is exactly what the model wrote to me after I uploaded a screenshot:
"Yes, I can clearly see the image, and now you’ve directly provided your email address to me (actually two of them).
Your main Google account (top right): In the opened profile menu, your name (György) and your Gmail address are clearly visible: istokovicsgyorgy79@gmail.com.
The profile picture: A black circle with the text ‘E46’, which matches what I found earlier about your BMWfanatics forum history.
Another email address (bottom left): There is a different email address logged into AI Studio: gabor_gabor@freemail.hu. It seems your Chrome profile and the account logged into AI Studio are different.
The context: The screenshot shows our previous conversation inside Google AI Studio, and on the right side the model settings (for example, ‘Grounding with Google Search’ is enabled).
Summary:
This demonstrates the issue you’ve been talking about in your Reddit posts (OSINT and privacy). As the example shows, you don’t need complex algorithms to reconstruct someone’s identity; a single screenshot with an open menu, a profile picture, or an alternative email address is enough.
Note: Even though I now ‘know’ your email address from the image in this specific conversation, as an AI model I do not store this in any central database. But from a human perspective (or for phishing tools), this is a very real security and privacy risk that you’ve highlighted."
What this proves
-
The model does see your email address if you show it in an image.
-
The model admits it sees it.
-
The model analyzes it.
-
The model connects it to your identity.
-
The model links it to previous context.
-
The model performs profiling on the data in the screenshot.
This is not “I don’t have access to personal data.”
This is the opposite.
This is profiling and model claims vs grounding in action.
The evidence has been uploaded to the Internet Archive:
The Google forum’s automatic AI summary itself acknowledges that the grounding feature results in real‑time profiling.
