Google quietly rewrote a promise it made three years ago — and the new wording means the emails and files you ask Gemini to read can end up in front of human reviewers.
In September 2023, when Google first let its chatbot reach into Gmail, Docs and Drive, it was explicit: content from those apps "is not seen by human reviewers" and would not be used to train the model. That sentence is still on Google's blog. But on October 16, 2026, a note was added to the company's help pages saying that for newer features, the same data "can be used to train generative AI models" and "can also be subject to human review in limited circumstances." There was no press release and no email to users — just a rewritten paragraph.
The change surfaced publicly this week after a GitHub project documenting the behaviour reached the top of Hacker News, drawing hundreds of upvotes and a wave of coverage. Google's own documentation now confirms the core of it.
What the new rules actually say
Gemini's Connected Apps help page, which covers Personal Intelligence — the feature that lets Gemini pull from Gmail, Drive, Docs, Calendar, Photos and Wallet on a personal Google account — says "a subset of Connected Apps data processed by Gemini is reviewed by human reviewers (including trained service providers)," outside contractors who grade Gemini's answers and write better ones.
What they review are the summaries, excerpts and inferences Gemini draws from your emails and files. Google says it does not train "directly" on your inbox or Drive, but it does train on those excerpts — and there is a catch in the fine print. When an email or file is short or especially relevant to the question, Google says "the summary may just be the email or file." In other words, the full text.
Reviewed data is kept for up to three years, disconnected from your account, and Google's help page says deleting your Gemini activity will not remove anything reviewers have already seen.
The claim Google has not confirmed
A separate allegation is circulating alongside the confirmed policy change. A GitHub user published a project titled "Documentation of Gemini File Leaking Behavior" claiming Gemini can pull Gmail and Drive content into a conversation on its own, without the user tagging a file with @Drive or @Gmail. Neither Google nor an independent tester has confirmed those reproduction claims, and the project's author is pseudonymous — but the report helped push the documented policy change into the open.
Who is affected — and how to opt out
The human-review rules apply to personal Google accounts. Work and school accounts play by different rules: Google says business and education content "is not human reviewed or otherwise used for Generative AI model training outside your domain without permission." Personal Intelligence is also unavailable in the European Economic Area, the UK, Switzerland and Nigeria, which limits exposure there.
Google's own page ends with a blunt warning: do not connect apps that hold confidential information you would not want a reviewer to see. Users who want out can disconnect Connected Apps from Gemini's Personal Intelligence settings, and turning off Gemini Apps Activity stops future data from being retained.
It is the second Gemini privacy row in a week — last week the company faced questions over training on users' chat history even on paid plans. The pattern is the same: the defaults quietly favour Google's data pipeline, and the only reliable protection is changing them yourself.
