Skip to content
sobolewski.proAI today, every eveningSubscribe

Once a day. Unsubscribe with one click. Privacy policy

Gemini privacy: humans may see your emails and files

Gemini privacy: Google added that some Gmail and Drive data connected to Gemini is reviewed by humans. What changed and how to turn it off.

4 min read

This article in other languages:Polski

In this article
  1. What exactly did Google change?
  2. What does Google say reaches the reviewers?
  3. What did the repository’s author show?
  4. Who does this affect?
  5. Is this new?
  6. Gemini privacy: how do you turn it off?
  7. FAQ

On November 9, 2026, the repository “Documentation of Gemini File Leaking Behavior” reached the front page of Hacker News (over 450 points). Its author points out that Google changed its rules: data from Gmail and Google Drive connected to consumer Gemini can now end up with human reviewers. When it comes to Gemini privacy, that is a 180-degree turn from Google’s 2023 promise.

What exactly did Google change?

In September 2023, when it launched Bard Extensions (Bard is today’s Gemini), Google wrote on its blog that our content from Gmail, Docs and Drive “is not seen by human reviewers”.

On October 16, 2026, an update was added to that same post. It says that data from Workspace apps may be used to train generative models and may also be reviewed by humans “in limited circumstances”.

The details are on the help page “About personalization with Connected Apps”. Google states plainly there that some connected app data processed by Gemini is reviewed by humans, including trained contractors. The reviewers assess whether Gemini’s responses were low quality, inaccurate or harmful.

The same page also advises against connecting apps that hold confidential information (quote: “Please don’t connect your apps to Gemini Apps if they contain confidential information”).

What does Google say reaches the reviewers?

According to the help page, among other things:

  1. Gemini does not train models directly on your Gmail inbox, Drive or Contacts, but it can use summaries, excerpts and inferences from your emails and files in its responses,
  2. before contractors review anything, Google disconnects conversations from your account and takes steps to limit personal data (e.g. differential privacy),
  3. humans don’t look at media from Google Photos, with exceptions (abuse, harm, submitted feedback),
  4. disconnecting an app does not delete data from Gemini Apps Activity.

Then there is the Gemini Apps Privacy Hub: conversations reviewed by humans are kept for up to three years, even after you delete your activity, because they are no longer linked to your account.

What did the repository’s author show?

The author describes a test on Gemini with Workspace connected (the Flash model with a high thinking level). It was enough to write something like “I wonder if I’m overpaying for internet”, without tagging @Gmail or @Drive, and Gemini went into the inbox on its own and pulled the email from the internet provider.

The result was slightly different each time. The author lists four variants: no retrieval at all, a general summary (e.g. a 122 USD bill from September 4), a few sentences with personal data and an account number, and the entire email including headers. In their view, automated personal-data filters can’t handle text like this, so all of it may reach reviewers in nearly raw form.

This is one person’s test, with no number of attempts and no logs, so let’s take it with a grain of salt. The rule that connected app data can reach humans, however, comes from Google’s official page.

Who does this affect?

This is about consumer Gemini, meaning personal Google accounts with connected apps turned on. In the Workspace privacy hub for business and education accounts, Google still states that “Your content is not human reviewed”.

In the comments, one user admitted that after checking their accounts they found the option turned on, and another saw it as one more reason to switch to models that run locally.

Is this new?

Similar questions come up with every assistant that is supposed to know our lives. I recently wrote about how Apple Intelligence builds a knowledge graph about us and the people close to us, except that there the data stays on the Mac. With Claude Personal Finance, we decide ourselves what to connect. What is new here is that Google walked back a clear promise, and the change is visible mostly in an addendum to a blog post and on help pages.

Gemini privacy: how do you turn it off?

Google gives you two options:

  1. Disconnect Google Workspace in the connected apps settings at gemini.google.com/apps. You lose responses based on your emails and files.
  2. Turn off Gemini Apps Activity or use temporary chats. According to the Privacy Hub, this stops future conversations from being reviewed, but it doesn’t undo the ones that have already been reviewed.

(Disconnecting an app doesn’t delete what is already in your Activity, so it’s worth reviewing that too.)

FAQ

Does Google read my Gmail through Gemini?

If Gmail is connected to consumer Gemini, Google’s help page says some connected app data may reach human reviewers, including contractors. If it isn’t connected, this change doesn’t affect you.

Does Gemini train on my emails?

Google says it doesn’t train models directly on your Gmail inbox and Drive, but in the October 16, 2026 update it allows Workspace app data to be used for training generative models.

Does this apply to Google Workspace business accounts?

According to the Workspace privacy hub, no: business customers’ content is not reviewed by humans or used for training outside their domain without permission.

How long does Google keep reviewed conversations?

Up to three years, even after you delete your Gemini Apps Activity.

  • AI
  • Google
  • Gemini
  • security

AI today, every evening

The day's biggest AI news and new posts from the blog, every evening. Short, with sources.

Once a day. Unsubscribe with one click. Privacy policy