Skip to main content
    Back to Blog
    14 min readLast reviewed:
    GDPR
    Gemini
    Google
    Europe
    AI Compliance
    Data Protection

    Is Gemini GDPR Compliant? A 2026 Guide for European Teams

    T

    Team Sonomos · Sonomos

    The Sonomos research team writes about AI privacy, data protection, and how to use generative AI safely at work.

    Short answer: Gemini is not GDPR compliant by default — but "Gemini" is three very different products with three very different GDPR postures. Consumer Gemini (gemini.google.com) has no Data Processing Addendum, may retain and human-review prompts, and is not suitable for processing EU personal data on behalf of an employer. Gemini in Google Workspace runs under Google's Cloud Data Processing Addendum, with the EU-U.S. Data Privacy Framework as the transfer mechanism and Standard Contractual Clauses as the fallback. Gemini Enterprise Agent Platform (formerly Vertex AI) comes under the same addendum through the Google Cloud terms and gives the most control, including an EU endpoint for supported models. This guide explains what each tier means under the GDPR in 2026, where the risks sit, and the controls that hold up under a supervisory-authority audit. Contract and residency details were checked against Google's pages on 18 November 2026.

    The three Gemini tiers and why they matter for GDPR

    The most common mistake European controllers make is treating "Gemini" as a single product. Google offers what are effectively three separate products sharing the same underlying models:

    Consumer Gemini (gemini.google.com, Google app) — a B2C product. Users sign in with personal Google accounts. Google's consumer Terms of Service apply, not an enterprise DPA. Prompts may be retained and reviewed by Google staff for quality and safety purposes unless the user opts out. Not covered by an enterprise data-processing agreement. Not suitable for processing personal data of EU data subjects on behalf of an employer.

    Google Workspace Gemini — the Gemini features built into Gmail, Google Docs, Google Sheets, Google Meet transcription, and the Workspace admin experience, plus the Gemini app used with a work account on a qualifying edition. Covered by the Cloud Data Processing Addendum, which a Workspace agreement incorporates or an admin accepts in the console (it used to be called the Data Processing Amendment). For transfers out of the EEA the addendum applies an "Alternative Transfer Solution" first and Standard Contractual Clauses only if Google has none, and Google's transfer page records that it has relied on Google LLC's Data Privacy Framework certification for EU and EEA data since 1 September 2023. Data regions cover Gemini prompts and responses at rest; where the model runs is covered only on some editions (see below).

    Gemini on Google Cloud (Gemini Enterprise Agent Platform, formerly Vertex AI) — the API-first product for developers building AI applications. Google's services summary now lists "Gemini Enterprise Agent Platform (formerly Vertex AI Platform)". There is no separate GDPR schedule to sign: section 5.2 of the Google Cloud terms says Google will only process customer data in accordance with the Cloud Data Processing Addendum. This tier gives the most control: zero-data-retention settings, and, per Google's data residency page, an EU multi-region endpoint that keeps model processing within EU member states (not the UK or Switzerland), while global endpoints "don't provide any data residency guarantees".

    Gemini tier-by-tier GDPR posture (2026)

    TierDPA available?Default trainingEU data residencySub-processors disclosedBest for
    Consumer GeminiNoOff by default; a subset of chats is human-reviewedNoLimitedPersonal use only
    Workspace GeminiYes: Cloud Data Processing AddendumNot without your permission or instruction (Workspace Service Specific Terms, section 12.11)At rest on editions with data regions; while processed only with Enterprise or Assured Controls Data RegionsYesTeams already on Workspace
    Agent Platform (formerly Vertex AI)Yes: Cloud Data Processing Addendum, through the Google Cloud termsExcluded by defaultEU multi-region endpoint for supported models; no guarantee on global endpointsYesAI product development

    The tier distinction is especially important because many organisations assume Workspace's DPA coverage extends to personal Gemini accounts employees open with their corporate Google credentials on a personal browser. It does not. A personal Google account is a consumer account, and Google's enterprise DPA does not extend to it regardless of what email address is used.

    Google's GDPR history and its implications

    Google has faced some of the largest and most substantive GDPR enforcement actions in Europe. Relevant history for 2026 deployments:

    • The CNIL (France) fined Google €50 million in January 2019 under the GDPR for transparency and consent failures in ad personalisation when users set up Android accounts. Its later fines against Google, including €150 million announced in January 2022 for making cookies harder to refuse than to accept, were imposed under French rules implementing the ePrivacy Directive rather than under the GDPR.
    • The Irish DPC, which is Google's lead supervisory authority under GDPR's one-stop-shop mechanism (Google's EU establishment is in Ireland), has issued multiple decisions and is the primary enforcement channel for complaints about Google's AI practices.
    • The EDPB's ChatGPT Taskforce report (May 2024) and its Opinion 28/2024 on AI models (December 2024) treat the training of a model as processing that needs its own lawful basis, and set out how authorities test legitimate interest and claims that a model is anonymous.

    For deployers, the implication is that Google's enterprise tiers have been shaped by regulatory engagement and come with mature data processing terms. Consumer tiers remain higher risk not because Google's practices are unusual, but because they are explicitly designed for a different legal relationship.

    What goes wrong in European Gemini deployments

    The patterns of failure are consistent across Google Workspace organisations:

    1. Employees open personal Gemini accounts at work. The most common breach pattern. An employee opens gemini.google.com, signs in with a personal Gmail account, and pastes a client email to summarise it. The employer has no DPA with Google for that account. The client's personal data is now processed under consumer terms with no Article 28 agreement. This is an Article 28 breach in a single prompt.

    2. Workspace Gemini enabled without confirming residency scope. An administrator assumes the EU data region covers AI processing. It covers Gemini prompts and responses at rest, but data in use only with Enterprise Data Regions (in Enterprise Plus and Frontline Plus, or as an add-on) or Assured Controls. On Business Standard, Business Plus or Enterprise Standard, stored data can sit in Europe while the model runs elsewhere, so check your edition before the DPIA says "EU only".

    3. Agent Platform projects without a configuration review. Engineering teams create Agent Platform (formerly Vertex AI) projects and build features that process customer data. The contract is not the gap, since the Cloud Data Processing Addendum applies through the Google Cloud terms. Configuration is: a global endpoint gives no residency guarantee, implicit caching is on by default until turned off, request-response logging writes prompts to BigQuery once enabled, and a customer billed outside Europe, the Middle East and Africa must certify that European data protection law applies to its transfers (section 4.2 of the addendum's European terms).

    4. No DPIA for Workspace Gemini rollout. Enabling Gemini across a Workspace organisation is a material change in processing. Article 35 requires a DPIA when processing is likely to result in high risk, and the UK ICO's guidance on AI and data protection says that "in the vast majority of cases" using AI will meet that threshold; the EU GDPR's Article 35 sets the same test. Google publishes a Gemini DPIA support guide describing its own processing, but a blanket "Gemini is part of our Workspace subscription, which has a DPA" does not substitute for a DPIA evaluating the specific use cases, data categories, and residual risks.

    5. Special-category data in Google Workspace Gemini. Gemini in Gmail can see email content. Gemini in Meet can transcribe conversations. If those emails or conversations involve health data, trade-union membership, or other Article 9 special-category data, the controller needs an Article 9(2) basis for the AI processing — not just for the original storage in Workspace.

    The control stack for a defensible deployment

    Contract layer

    For Workspace Gemini: Confirm your Workspace agreement incorporates the Cloud Data Processing Addendum. Google's admin help says to opt in only if it does not, and to opt in when unsure because doing so "won't make any difference" if it already does; a super administrator does it under Account settings > Legal and compliance, where you also name your supervisory authority. Section 12.11 of the Workspace Service Specific Terms bars Google from training its generative models on your data without permission. Make sure sub-processor notices reach a monitored mailbox.

    For Gemini Enterprise Agent Platform (formerly Vertex AI): There is nothing extra to sign, because section 5.2 of the Google Cloud terms already ties Google's processing to the Cloud Data Processing Addendum. The work is configuration: choose an EU endpoint for a model that supports it, turn off implicit caching and leave request-response logging off where the workload warrants zero retention, and review the sub-processor list for the regions you use.

    For both: Google applies the Data Privacy Framework to restricted transfers to the US, and its Standard Contractual Clauses take over if it stops relying on the framework, with nothing new to sign. Record both in the transfer impact assessment with the date you checked Google's certification. Which Gemini product falls under which Google addendum, and where an admin accepts it, is set out in our Gemini DPA guide.

    Tier-and-settings layer

    • Disable or restrict consumer Gemini access across managed devices. Workspace admin controls can block personal Google accounts on organisational devices, which is the most direct way to keep staff off gemini.google.com under personal accounts.
    • Confirm training exclusion for Workspace and Agent Platform workloads in admin settings.
    • Set the Workspace data region to Europe and document whether your data regions version covers data in use, not only at rest.
    • For Agent Platform: configure zero-data-retention where available and appropriate, and pin workloads to EU endpoints.
    • Review and configure the Gemini app admin settings in the Workspace console — which Gemini features are enabled, which data sources they can access (Drive, email), and what audit logging is active.

    Data-minimisation layer

    The transfer risk is most directly addressed by minimising personal data in the prompt. Locke 1.0 detects personal identifiers — names, email addresses, account numbers, health terms — on-device and replaces them with one-way placeholders before submission, addressing the Article 32 pseudonymisation obligation and providing the supplementary measure referenced in EDPB Recommendations 01/2020. For Agent Platform (formerly Vertex AI) applications, this pseudonymisation step belongs in the application layer before the API call is made; reversible tokenization is planned shortly after 1.0.

    Governance layer

    • Article 30 register updated for each Gemini tier in use, with lawful basis, recipients, transfer mechanism and retention period. For Workspace and Google Cloud the recipient is your Google contracting entity, set by billing address on Google's contracting entity page: Google Cloud EMEA Limited for most of Europe, the Middle East and Africa, or Google Cloud France, Google Cloud Italy S.r.l. or Google Cloud Poland Sp. z o.o. for customers billed there, with Google LLC among the sub-processors. Google Ireland Limited provides the consumer Gemini app in the EEA and decides how its chats are used, so it cannot be your processor.
    • DPIA completed for high-risk uses and refreshed when the Gemini feature set materially changes (which has happened frequently as Workspace Gemini has expanded).
    • Transparency notices updated to identify Gemini as an AI tool, Google as a recipient, and the transfer mechanism.
    • Data subject rights procedure documented for Gemini processing — how do you respond to an erasure request when the data subject's information was included in a Gmail summary, a Meet transcript, or an Agent Platform prompt?

    EU AI Act overlay

    Google is a General-Purpose AI (GPAI) provider under the EU AI Act, and Gemini's family of models carries GPAI obligations from August 2025: transparency about training data, technical documentation, model evaluation, and serious-incident reporting. Deployers using Workspace Gemini or Agent Platform should verify Google's GPAI compliance documentation in their vendor risk management process.

    For high-risk uses (employment tools, credit screening, educational scoring, law enforcement support), the Annex III obligations apply to the deployer. Using Gemini on Agent Platform to power an employment-decision or credit-assessment tool makes the controller a high-risk deployer, with duties of human oversight, monitoring and record-keeping.

    Workspace Gemini features that summarise HR records or assist in hiring decisions sit at the boundary of the high-risk classification, so document whether each use case falls within Annex III. Regulation (EU) 2026/1744, published in July 2026, moved the Annex III date from 2 August 2026 to 2 December 2027.

    Frequently asked questions

    Is Google Workspace Gemini GDPR compliant?

    Workspace Gemini has the required contractual infrastructure: the Cloud Data Processing Addendum, a transfer mechanism (the Data Privacy Framework, with Standard Contractual Clauses as the fallback), 30 days' notice of new sub-processors, and a bar on training Google's generative models on Workspace data without permission. The deployer is still responsible for lawful basis, transparency, DPIA where required, data subject rights, and Article 32 measures, and should check whether its data regions version covers data in use. Under the addendum, with a completed DPIA and appropriate access controls, it is a defensible deployment.

    Can EU teams use consumer Gemini for work?

    No. Consumer Gemini (gemini.google.com or the Google app with a personal account) is not covered by an enterprise DPA and is not suitable for processing personal data of EU data subjects on behalf of an employer. Employers should block consumer Gemini access on managed devices and provide a sanctioned alternative (Workspace Gemini or Agent Platform).

    Does Google use Workspace Gemini prompts to train its models?

    Not without your permission. Section 12.11 of the Workspace Service Specific Terms says Google will not use customer data to train or fine-tune the generative models behind Workspace's AI services without the customer's prior permission or instruction. Agent Platform prompts are similarly excluded by default. Consumer Gemini may use conversation data, including through human review, depending on the user's activity settings. These terms can change, so check them again when you write the DPIA.

    Do we need a DPIA before enabling Gemini in Google Workspace?

    Almost certainly yes for organisation-wide rollout involving personal data. The UK ICO's AI guidance says that in the vast majority of cases using AI is likely to be high-risk processing that requires a DPIA, and Article 35 of the EU GDPR applies the same test. Google's Gemini DPIA support guide describes Google's side of the processing, not your purposes or lawful basis. The DPIA should evaluate the Gemini features enabled, the personal data they can reach (email, calendar, Drive), the lawful basis, the transfer mechanism, and residual risks.

    What is the GDPR position on Gemini's ability to access our Google Drive and Gmail?

    When Gemini in Workspace accesses Gmail or Drive to answer a query, that access is a processing activity under the GDPR. The controller's lawful basis for storing that data (most commonly legitimate interest or contract performance) extends to AI-assisted access, provided the access is consistent with the purpose for which the data was collected and disclosed in the privacy notice. The issue arises when the AI access creates new processing purposes — for example, using email patterns to infer employee behaviour — that were not disclosed to data subjects.

    The bottom line

    Consumer Gemini is not suitable for EU personal-data processing in a business context. Workspace Gemini is the right default for organisations already on Google Workspace: the Cloud Data Processing Addendum and a transfer mechanism are in place, training exclusion applies, and the admin controls are maturing. Gemini on Agent Platform (formerly Vertex AI) gives the most control for custom AI development with EU data.

    Unless you have configured processing to stay in the EU (Enterprise Data Regions or Assured Controls in Workspace, or an EU endpoint on Agent Platform for a model that supports it), plan on prompts being processed outside the EEA under the Data Privacy Framework, with Google's SCCs behind it. The answer is not to avoid Google's AI stack; it is to pseudonymise personal data before it reaches the model, document the transfer mechanism, and complete the governance steps that make the posture auditable.

    Send the AI less personal data

    Locke runs on your own computer and replaces the names, email addresses and other personal data it detects with placeholders before a prompt reaches the AI tools it screens, so less personal data is transferred to the AI provider. It does not make an AI tool GDPR compliant or replace a DPA, a transfer mechanism or a DPIA, and anything it does not detect is sent as typed.

    Sonomos, which publishes this blog, makes Locke. Free for one person on x64 Windows and Linux; macOS is coming soon.

    This article is general information about privacy and AI, not legal advice, and reading it does not create a solicitor–client or attorney–client relationship. Regulatory requirements turn on your specific facts and jurisdiction and change over time. Consult qualified counsel before relying on anything here, and check the publication and last-reviewed dates above. We may revise posts, but we do not undertake to keep any post current as the law changes.

    Reference pages for this post