news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

IT Home reported on July 15 that according to a report by X user @Kevin Bankston, Google's artificial intelligence service Gemini AI seems to read private documents in Google Drive without the user's explicit permission.

IT Home noticed that Bankston described the problem in detail in his tweet. Initially, he found that Gemini seemed to be reading his Google Drive documents.But I can't find an option to disable this feature in the relevant settingsUpon further investigation, he discovered that this was a problem with Google Drive itself, not Google Docs, but that both may have similar vulnerabilities.

Moreover, according to Gemini AI, the privacy settings for controlling information sharing should be manageable by users, but Bankston could not find them at all. This either means that Gemini is "hallucinating" (i.e. providing false information) or there is a problem with Google's internal servers. In either case, this makes users question the security of their data, even if Google claims that the scanned data will not be used to train Gemini AI.

Bankston eventually found the relevant settings in a completely different location.But even with that disabled, Gemini still automatically scans all documents of the same type (in this case, PDFs) after he clicks on one.Bankston speculated that this may be due to the Google Workspace Labs feature he enabled in 2023, which may have overwritten the established Gemini settings.

Even if the issue is limited to Google Workspace Labs users, it is still a pretty serious negative impact for users who help Google test its latest technologies. Especially when sensitive information is involved, the user's right to informed consent is very important.