Gemini can show your staff's emails to human reviewers. Here is the business fix

If your employees connect a personal Gemini account to their Gmail and Drive, some of what Gemini reads there can end up in front of a human reviewer. That is not a leak. It is what Google's own help page on Connected Apps now says: "Data from Connected Apps is used and human reviewed per the Gemini Apps Privacy Notice."

The page adds that Gemini creates summaries and excerpts of emails and files, that these are used to train Google's models, and that "where the email or file is brief or particularly relevant", the summary "may just be the email or file".

The short version

• In the consumer Gemini app, content pulled from connected Gmail and Drive can be used for training and reviewed by people, according to Google's help page.
• The risk for a business is not Google. It is staff using personal AI accounts on company email, quotes and contracts.
• Decide which AI tools may touch company data, on which accounts, before an employee decides for you.

What a user found, and what Google confirms

The story reached the top of Hacker News through a write-up on GitHub by a user who tested the consumer Gemini app with Connected Apps switched on. The author argues that Google had promised in 2023 that personal Workspace content would not be seen by human reviewers, and that this promise has since changed.

The author also reports that a casual question, such as wondering whether an internet bill is too high, sometimes made Gemini pull a matching email on its own, from a short summary up to the full text with account numbers. Those are one person's tests, and results varied from try to try. The part that matters most for a business is not in dispute, because it comes from Google: connected email and files can be summarised, used for training and reviewed by people.

Why this is a company problem

Picture an illustrative sales team of eight. Each person uses a free AI assistant on their phone, logged into the same Google account they use for work email. One asks it to "summarise the quote I sent to the hotel in Marrakech". Another asks for help replying to a supplier about a late payment.

Nobody did anything dramatic. Yet client names, prices, contract terms and payment details now flow through consumer services under consumer privacy terms, not the terms your company would sign for business use. Multiply that by every team and every month.

Business editions of AI tools usually come with different data terms than consumer apps. Check the terms of each product you allow, in writing, rather than assuming.

A one-page AI data rule

• Accounts: company data only in AI tools on company accounts, with business terms you have read.
• Connections: no connecting personal AI assistants to work email, drives or calendars.
• Content: never paste ID documents, contracts, payroll or client lists into a consumer AI app.
• Settings: for each approved tool, note who switched off training and history settings, and when.
• Exceptions: a named person approves anything outside these rules.

Tidy access before you add AI

AI assistants read whatever the account they run under can read. If contracts, HR files and client records sit in shared folders that half the company can open, any AI tool connected to those accounts can read them too.

That is why access control comes before AI. Holol-GED gives sensitive documents a secure, searchable home at scale, so files live behind clear permissions instead of personal drives and forwarded attachments. When you do add AI on top, as we do in our AI agent projects, you decide exactly what it may read.

Read the help page, not the headline

Every AI tool your team uses has a page like Google's. Ten minutes reading them is cheaper than finding out later where your clients' data went.

Sources

• Google Gemini Apps Help: About personalization with Connected Apps
• GitHub: Google Gemini routes personal Drive files and emails to human annotators (Hacker News discussion)