Google's Gemini Found Reading PDF Files in Drive Without User Consent, Unclear if Due to a Bug

Google's artificial intelligence application, Gemini, can access various types of information stored by users in Google, including Google Docs and Sheets, provided it has the users' authorization. However, this authorization does not extend to files stored in Google Drive.

On social media platforms, a user named Kevin Bankster noticed that Gemini was able to read PDF documents in Google Drive, which could potentially involve private content that should not be accessible to Gemini.

According to Google's statement, Gemini does not use users' private data to train artificial intelligence. Despite this, the unauthorized reading of users' private files still poses a risk of privacy breach.

Kevin Bankster contacted Google support regarding this issue, but Google's support staff were also unclear about the problem at the time. A reasonable guess is that some internal system failure on Google's servers might have allowed Gemini to access unauthorized files.

The bigger issue is that whether or not users utilize Gemini, as long as their data is stored on Google Drive or other Google services, Gemini might be able to access it, which completely contradicts Google's privacy statements.

This situation is highly risky for both Google and its users. Later, Kevin Bankster found Gemini's permission settings in a hidden location, but the settings page also showed that Gemini's access to Gmail, Google Drive, and Google Docs was disabled, implying that Gemini did indeed read files without authorization.

Currently, Google has not issued a response. It is highly probable that this issue is due to some sort of bug, but it is still unclear whether user data has been used to train models or given to humans for review.