The Gemini neural network was caught scanning PDF files of Google Drive users

The rapid development of neural networks has long raised concerns among experts about ensuring privacy when interacting with such algorithms. This week there was further confirmation that these fears are not in vain. Social network user X Kevin Bankster drew attention to the fact that Google’s Gemini neural network automatically scans PDF files stored in Google Drive without obtaining the user’s explicit consent.

Image source: Google

«Just opened my tax return in Google Docs and Gemini summarized it without permission. So… Gemini automatically scans even personal documents I open in Google Docs? I didn’t ask for this. Now I need to find new settings that were never communicated to me to disable this,” Bankster wrote on his Network X account.

What exactly caused Gemini to behave this way is not yet clear. According to the algorithm itself, the privacy settings used to inform Gemini should be publicly available, but this is not the case. This could mean that either the algorithm is producing false information, or that some of Google’s internal systems are clearly not functioning correctly. In any case, this does not look very good, even if we assume that the neural network does not use users’ personal data for training.

As for Bankster, after some time he wrote that he was able to find an option in the settings that, when activated, allows Gemini to scan documents in Gmail, Google Docs and Google Drive. At the same time, he noted that the option was disabled, but for some reason the algorithm still scanned the document and created a generalized text based on it. Bankster believes the issue may be related to the fact that he signed up for Google Workspace Labs in 2023, which may have caused Gemini’s settings to not apply correctly.