Is business data safe in Gemini?

What you will be able to do

You will separate consumer Gemini Apps data handling from Google Workspace Gemini, list inputs that stay out of prompts, and cite Google’s help instead of a compliance badge.

The honest answer

Risk is the product of four inputs: what staff submit or connect, which account class they use, the current vendor terms, and Friday behavior. The interface does not sanitize a bad paste.

Personal Gemini Apps

Scope is the Gemini Apps Privacy Hub: gemini.google.com, the Gemini mobile apps, and the other consumer surfaces that notice lists. Stated uses include providing and improving services, extending to the generative models that power Gemini Apps. Human reviewers see a subset. The notice tells users not to submit confidential information they would not want reviewed or used to improve machine-learning technologies.

Gemini Apps Activity, including Keep Activity, controls whether data is used to improve Google AI, including those models. With Keep Activity off, Google still uses chats to respond and to protect users, including human review, and states that Gemini settings do not govern anonymized processing for service improvement. Chats that reviewers have seen can be retained for up to three years even after you delete activity. Read the retention section before you write policy. Do not describe a single toggle as if it erased the chats.

Consumer Connected Apps (Gmail, Drive, and others) are processed under that same notice. Workspace-class Connected Apps are unavailable when Keep Activity is off. A personal login pointed at the shop inbox is not Workspace Gemini. See Workspace file access.

The hub directs work and school accounts to different terms.

Google Workspace

The Workspace generative AI privacy hub states that prompts and Workspace content are not used to train generative models outside the domain without permission, and are not human-reviewed for that purpose. The Gemini app admin article splits qualifying Workspace editions (Workspace terms) from users without a qualifying edition (consumer Gemini terms; avoid confidential input).

Re-read both documents before policy is frozen. They move.

What not to paste

Align with customer data rules:

• Payment credentials and passwords
• Unnecessary health detail and full charts
• Government IDs, children’s data, unrestricted HR files
• Access codes and other shared-email secrets

Submit role plus required facts. Redact identifiers the draft does not need.

Twins: Claude, ChatGPT.

What this article is not

Not legal advice, and not an attestation that the practice holds any certification. Workspace privacy commitments are Google’s product terms. They do not automatically extend coverage to the clinic. Confirm the executed agreement and counsel.

Try this job: a one-page Gemini data rule

1. Red class: never submit.
2. Yellow class: Workspace only, admin-approved, minimum fields.
3. Green class: public copy, templates, anonymized examples.
4. Assign which roles may use a personal Google AI login for work.
5. Re-review when either privacy hub changes.

Short close

Minimize inputs, match the account class to the sensitivity, and cite the current Google hubs.