Is Gemini safe for work?
16 min read
Is Gemini safe for work? A dated, sourced table of what Google does with company data in the Gemini app, Workspace and the Gemini API, and a verdict for each.
It depends on which Gemini. On a work account with a qualifying Google Workspace edition, Google does not train on your chats or send them to human reviewers, and Gemini is safe for most company data. On a personal Google account, free or paid, chats train Google's models and people read a sample of them. The short answer, as of 15 November 2026:
Product | Safe for company data? | Why |
|---|---|---|
Gemini app on a personal Google account (free, Google AI Plus, Pro, Ultra) | No | Chats train Google's models by default and a subset goes to human reviewers. Reviewed chats are kept for up to three years, even after deletion. No admin, no record of what was pasted. |
Gemini app on a work account without a qualifying edition | No | An additional service under the consumer terms. Google tells admins to warn these users not to enter confidential information. |
Gemini app and Gemini in Workspace on a qualifying edition | Yes, for most work | No training and no human review without your permission. Admin-set retention, Vault, audit logs, sharing controls. Data regions are the US or the EU, with no UK option. |
Gemini API, unpaid quota | No, outside the UK, EEA and Switzerland | Prompts and responses improve Google's products and reviewers may read them. In the UK, EEA and Switzerland the paid terms apply to all use. |
Gemini API paid quota and Gemini Enterprise Agent Platform (formerly Vertex AI) | Yes, with configuration | No training. Abuse logs for a limited period. Zero data retention is self-serve on Agent Platform only, and needs several settings. UK processing for some models only. |
Everything below is drawn from Google's own pages, read on 15 November 2026 and linked at the end. Two names have changed: Vertex AI is now Gemini Enterprise Agent Platform, and NotebookLM became Gemini Notebook on 16 July 2026. NotebookLM has different terms, so it has its own guide: Is NotebookLM safe for work?
What Google does with your data, by product
Gemini app, personal account | Gemini app and Gemini in Workspace, qualifying edition | Gemini API (Google AI Studio) | Gemini Enterprise Agent Platform (Vertex AI) | |
|---|---|---|---|---|
Terms that apply | Google Terms of Service and the Gemini Apps Privacy Notice. Provided by Google LLC in the UK and US | Your Workspace agreement and the Cloud Data Processing Addendum | Gemini API Additional Terms, effective 23 March 2026 | Google Cloud terms and the Cloud Data Processing Addendum |
Trains on your content by default | Yes, while Keep activity is on. It is on by default for adults | No, unless you give permission or instruction | Unpaid: yes. Paid: no. UK, EEA and Switzerland: no, on either | No, unless you give permission or instruction |
Human review | Yes. A subset of chats, disconnected from the account first. Reviewed chats are kept for up to three years and survive deletion of your activity | No. Feedback is voluntary, does not train models, and may be kept for up to 18 months | Unpaid: reviewers may read and annotate input and output. Paid: logged only to detect abuse and meet legal duties | Only prompts flagged by abuse classifiers, for customers on the online terms. A Cloud Master Agreement exempts you by default |
How to turn training off | Turn off Keep activity, or use a temporary chat. Feedback still shares the last 24 hours of chats | Not needed | Use a project with Cloud Billing enabled | Not needed |
How long chats are kept | 18 months by default. Options: 3 or 36 months, or no auto-delete. Keep activity off or a temporary chat: 72 hours | Gemini app: 18 months by default; admins pick 3, 18 or 36 months, or off (72 hours). Side panel in Docs, Gmail and others: 90, 540 or 1,080 days, or indefinite. Vault retention rules override user deletion and temporary chats | Paid abuse logs: a limited period, with no number published. Developer logs: 55 days by default, on by default for Interactions API calls. Grounding with Google Search or Maps: 30 days | Flagged prompts: up to 90 days in your project's region. Gemini 4 Carbon: prompts and responses for up to 30 days inside your environment. Implicit cache: up to 24 hours |
Zero data retention | No | No | Not guaranteed. Google logs paid-tier prompts and responses for a limited period for abuse monitoring, and says to use Agent Platform if you need guaranteed zero data retention. You can cut the rest: set | Self-serve, by similar settings. Not possible with Grounding with Google Search or Maps, or the Deep Research agent |
Where it is stored and processed | No region setting | Data regions: United States or Europe (the EU). No UK region. Storage at rest is covered from Business Standard up; Business Starter is not supported. In-region processing needs Enterprise Data Regions (Enterprise Plus, Enterprise Essentials Plus, Frontline Plus, or the add-on) | Paid data may be held briefly or cached in any country where Google has facilities | At rest in the location you choose. Processing follows the endpoint; the global endpoint gives no guarantee. The EU multi-region excludes the UK. UK (europe-west2) processing: Gemini 4 Argon, 3.5 Flash and 2.5 Flash. Gemini 4 Carbon and Hydrogen: US and EU only |
Admin controls | None | On or off by organisational unit or group. Gemini features per service (Gmail, Drive, Docs, Meet, Chat). Temporary chats and user deletion. Gemini in Chrome through the GeminiSettings policy. AI control center on Enterprise Standard and Plus. No control specific to the mobile app | Project owners turn logging on or off | Request-response logging is off by default. Abuse logs follow Data Residency, Access Transparency and VPC Service Controls, but not customer-managed keys |
Audit trail | None | Gemini audit logs in the Reporting API and the audit investigation tool, for all Workspace customers. Vault can search and export Gemini app prompts and responses | Your own developer logs | Your own request-response logs in BigQuery, if enabled |
Sharing | Public links. Anyone with the link can read the chat | Public links are off by default. Sharing through Drive is on by default and follows your Drive settings | Not applicable | Not applicable |
Certifications | None. Google's compliance table shows no HIPAA, SOC or ISO coverage under the consumer terms | SOC 1, 2 and 3; ISO 27001, 27017, 27018, 27701, 9001 and 42001; FedRAMP High; BSI C5:2020; HIPAA support. Gemini in Chrome is excluded | Not stated on the pages read | Not read for this guide |
Price (US) | $0; Google AI Plus $4.99; Pro $19.99; Ultra $99.99 or $199.99 a month | Included in Workspace: Business Starter $7, Standard $14, Plus $22 per user a month on a one-year commitment. Enterprise: contact sales | Gemini 4 Argon and Carbon: $2 in, $10 out per million tokens until 30 November 2026, then $4 and $20. Gemini 4 Hydrogen: $1.50 and $7.50, with a free tier | Usage-based. Price list not read for this guide |
Gemini Enterprise, from $21 per seat a month, is a separate agent and search product on Google Cloud and is outside this guide.
What that means for your staff
On a personal Google account, every chat is saved, used to train models and open to human review unless the person has turned Keep activity off. Paying for Google AI Pro or Ultra changes the limits and the models. The privacy notice is the same. Nobody at your company can see what was pasted or delete it, and a chat a reviewer has seen stays with Google for up to three years.
On a work account without a qualifying edition, such as Business Base, Essentials Starter, Workspace Individual, legacy G Suite or Cloud Identity, the Gemini app is an additional service. The address is the company's and the terms are the consumer ones.
On a qualifying Workspace edition, the position is close to what you already accept for Gmail and Drive. Prompts are customer data under the Cloud Data Processing Addendum. Admins set retention, Vault can export conversations, and the audit log shows who used which Gemini feature in which app. Gemini retrieves only files and mail the user can already open, so every overshared file in Drive is now one prompt away.
On the API, billing decides it. A key from a project without billing sends prompts into product improvement, except in the UK, EEA and Switzerland. A key from a billed project does not, but the Developer API makes no residency promise. Agent Platform has the regional endpoints.
The red list: what never goes into any Gemini
Put this list in the policy and in the onboarding deck. It applies on every product, Workspace included.
Passwords, API keys, tokens, connection strings, private keys
Customer personal data: names with contact details, account numbers, health or financial records
Source code that is a trade secret, or anything under a customer NDA
Unreleased financial results, deal terms, board papers
Anything legally privileged
Patient or health data, unless you have a signed BAA with Google, and never in Gemini in Chrome
Security incident details while the incident is open
If staff need AI for work on this list, they need a tool that cannot send the data to a destination you have not approved. That is the last section.
The risks that stay, whatever product you buy
The wrong account. A browser can hold a work and a personal Google account side by side. The same gemini.google.com page serves both, under different terms.
Prompt injection. A document, email or calendar invite that Gemini reads can carry instructions. GeminiJack, found by Noma Security and reported on 10 December 2025, affected Gemini Enterprise and Vertex AI Search. A hidden instruction in a shared document, email or invite made Gemini search for sensitive data and place it in an image URL the attacker controlled. No click was needed. In January 2026 Miggo Security showed that an instruction hidden in a calendar invite made Gemini copy summaries of private meetings into a new event when the user asked about their schedule. Google changed its systems after both reports. Each connected source is another input an attacker can write to.
Connected and third-party apps. Google states that it does not monitor or secure data from custom third-party connected apps, and the Workspace commitments do not cover third-party apps used with Gemini.
Gemini in Chrome. It sends Google the URL and page content of the tabs a user selects. It is available by default to users with Gemini access, today those signed in to Chrome in the US. It has no SOC, ISO, FedRAMP or BSI C5 coverage, and Google blocks it for customers who have signed a BAA.
Public links. A shared chat is readable by anyone who has the link. Deleting the chat does not delete the link.
Wrong answers. Gemini states false things fluently, on every product. Nothing goes to a customer, a regulator or production without a person who understands it checking it.
Settings to change today
Every personal account used for work (until you move people off them):
Open Gemini Apps activity and turn off Keep activity.
Use temporary chats for anything else, and still assume 72 hours of retention.
Do not send thumbs up or down feedback.
Open Settings, then Your public links, and delete old links.
Workspace (Admin console, Generative AI):
Confirm your edition gives the Gemini app as a core service. If it does not, turn the app off.
Gemini app, Gemini conversation history: choose 3, 18 or 36 months, and decide whether users may delete chats or start temporary chats.
Gemini app, Sharing: leave link sharing off.
Gemini for Workspace, Conversation history and deletion: set auto-deletion for the side panel.
Data, Compliance, Data regions: choose the United States or Europe, and record that the UK is not an option.
Set Vault retention rules for Gemini if you hold other Workspace data for legal reasons.
Set the GeminiSettings policy for Gemini in Chrome.
Fix overshared Drive folders before a wide rollout.
Gemini API and Agent Platform:
Use only keys from billing-enabled projects.
To keep stored data to a minimum, set
storeto false, leave session resumption off, and avoid grounding with Google Search and Maps. The Gemini API still logs paid-tier prompts for abuse monitoring for a limited period. For guaranteed zero data retention, use Agent Platform.For UK processing, use Agent Platform with the europe-west2 endpoint and a model that supports it. Do not use the global endpoint.
What to put in the acceptable use policy
Six lines cover it. Which Google accounts are approved for Gemini, and that personal accounts are banned for work. The red list above. Who may turn on connected apps. That outputs are checked before they leave the building. Where to report a paste that should not have happened, without blame. The tool to use for work on the red list.
Our guide to shadow AI and acceptable use policies has a policy outline. The other tools your staff use are in the AI tools at work hub.
Keep sensitive work on approved models
The Gemini app cannot be pointed at a model you run. Workspace settings change what Google keeps, and every prompt is still processed by Google. For the red list, the decision is which tool to use and where its requests may go.
Aventail™ Connect configures supported AI apps to send their model requests through Aventail™ Link, an inference engine and gateway that runs on each machine, under routing rules set in Aventail Control. Claude Desktop, Claude Code, OpenCode and Codex are supported today. Link evaluates the rules on the device. When no permitted destination exists, the request fails with an error. Hosted Control receives management, device and routing events, and prompts and responses do not pass through it.
Google's models can be one of the permitted destinations. Google Gemini is a provider in Control's API key vault, used through your own Google account, and Aventail™ supports Gemini 4 Argon and Carbon. A Gemini model reached this way still runs at Google, under the API terms above.
In practice that gives you two lanes:
Gemini on a qualifying Workspace edition for general work, with the settings above.
An approved assistant routed by Aventail™ for the red list: a model on the person's own machine with no cloud fallback, and Gemini through your own API account only for work your policy allows.
The free plan covers three connected devices. The rules are explained in routing between local models and cloud APIs.
Frequently asked questions
Does Gemini use my data for training?
On a personal account, yes, while Keep activity is on. On a qualifying Workspace edition, no. On the Gemini API, unpaid use is used for product improvement outside the UK, EEA and Switzerland, and paid use is not.
Can Google employees read what we type into Gemini at work?
On a qualifying Workspace edition, Google states that chats and uploaded files are not reviewed by people without your permission. On a personal account, or a work account where Gemini is an additional service, a subset of chats is reviewed.
Is Gemini safe for confidential information?
On a personal account, no. On a qualifying Workspace edition, yes for information you already trust to Gmail and Drive, within the retention you set. Information on the red list above does not go into the Gemini app on any plan.
Is Google AI Pro safer than the free Gemini app?
No. Google AI Plus, Pro and Ultra are personal plans under the same Gemini Apps Privacy Notice as the free app. They add models, limits and storage.
How long does Google keep Gemini conversations from work accounts?
In the Gemini app, 18 months by default. An admin can choose 3, 18 or 36 months or turn history off, which leaves 72 hours. If the organisation uses Vault for Gemini, Vault retention rules apply even to deleted and temporary chats.
Is Gemini HIPAA compliant?
Google says the Gemini app and Gemini in Workspace can support HIPAA workloads on qualifying editions under its Business Associate Agreement. Gemini in Chrome is not covered and is blocked for BAA customers. Consumer accounts are not covered.
Does Google store Gemini data in the UK?
Not through Workspace. Data regions offer the United States or Europe, meaning the EU. On Agent Platform, the europe-west2 endpoint processes requests in the UK for Gemini 4 Argon, Gemini 3.5 Flash and Gemini 2.5 Flash, as listed on 15 November 2026.
Related guides
Is NotebookLM safe for work?
Sources
All read on 15 November 2026.
Google, Gemini Apps Privacy Hub (notice updated 13 October 2026): https://support.google.com/gemini/answer/13594961
Google, Manage and delete your activity in Gemini Apps: https://support.google.com/gemini/answer/13278892
Google, Share your chats from Gemini Apps: https://support.google.com/gemini/answer/13743730
Google Workspace, Generative AI in Google Workspace Privacy Hub (updated 6 October 2026): https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub
Google Workspace, Turn the Gemini app on or off: https://knowledge.workspace.google.com/admin/generative-ai/gemini-app/turn-the-gemini-app-on-or-off
Google Workspace, Manage Gemini in Workspace conversation history settings: https://knowledge.workspace.google.com/admin/generative-ai/workspace-with-gemini/manage-gemini-in-workspace-conversation-history-settings
Google Workspace, Control temporary chats and chat deletion: https://knowledge.workspace.google.com/admin/generative-ai/gemini-app/enable-temporary-chats-in-gemini-app
Google Workspace, Turn conversation sharing on or off: https://knowledge.workspace.google.com/admin/generative-ai/gemini-app/turn-conversation-sharing-on-or-off
Google Workspace, Explore the AI control center: https://knowledge.workspace.google.com/admin/generative-ai/explore-the-ai-control-center
Google Workspace, Choose a geographic location for your data: https://knowledge.workspace.google.com/admin/compliance/choose-a-geographic-location-for-your-data
Google Workspace, Data covered by data regions: https://knowledge.workspace.google.com/admin/compliance/data-covered-by-data-regions
Google Vault, Use Vault to search Gemini app: https://knowledge.workspace.google.com/vault/search/use-vault-to-search-gemini-app
Google Workspace Updates, Gemini audit logs in the Reporting API and investigation tools (July 2025): https://workspaceupdates.googleblog.com/2025/07/gemini-audit-logs-reporting-api-audit-and-security-invesitgation-tools.html
Google Workspace pricing: https://workspace.google.com/pricing
Google, Gemini subscriptions: https://gemini.google/subscriptions/
Google, Gemini API Additional Terms of Service (effective 23 March 2026): https://aistudio.google.com/docs/terms
Google, Zero data retention in the Gemini Developer API: https://aistudio.google.com/docs/zdr
Google, Data logging and sharing: https://aistudio.google.com/docs/logs-policy
Google, Gemini Developer API pricing: https://aistudio.google.com/docs/pricing
Google Cloud, Gemini Enterprise Agent Platform and zero data retention: https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention
Google Cloud, Abuse monitoring: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring
Google Cloud, Data residency: https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/data-residency
Google Cloud, Gemini Enterprise Agent Platform name changes: https://docs.cloud.google.com/gemini-enterprise-agent-platform/vertex-ai-name-changes
Google, NotebookLM is now Gemini Notebook (16 July 2026): https://blog.google/innovation-and-ai/products/gemini-notebook/notebooklm-gemini-notebook/
Infosecurity Magazine, Google fixes Gemini Enterprise flaw (10 December 2025): https://www.infosecurity-magazine.com/news/google-fixes-gemini-enterprise-flaw
TechBriefly, Google patches Gemini flaw that turned invites into attack vectors (21 January 2026): https://techbriefly.com/2026/01/21/google-patches-critical-gemini-flaw-that-turned-invites-into-attack-vectors/
Aventail™, Connect overview: https://docs.aventail.co.uk/docs/connect/overview
Aventail™, Routing: https://docs.aventail.co.uk/docs/fleet/routing
Aventail™, Trust: https://aventail.co.uk/trust