Copilot 365 vs Google Workspace AI: compliance comparison for enterprise buyers

TL;DR. Both products are tenant-bound enterprise AI: customer data stays in the buyer's existing Microsoft 365 or Google Workspace tenant by default, no training on customer content, and a BAA available to healthcare buyers (included by default in Microsoft's DPA for covered entities and business associates; at Google, accepted by a Workspace administrator in the Admin console on any edition except the legacy free edition, with Gemini Notebook and Gemini in Chrome outside it). The compliance picture diverges on foundation-model subprocessors (Copilot added Anthropic as a subprocessor in January 2026, outside EU Data Boundary scope, and OpenAI-operated models in June 2026), on how each vendor scopes data residency (Microsoft's EU Data Boundary, with exceptions for flex routing at peak demand, Anthropic models and Bing web search queries; Google's Workspace Data Regions, which cover Gemini prompts and responses but not Gemini Notebook), and on the AI Act deployer angle for HR-adjacent Copilot Agents.

How the two compare

The full read for each lives in the Microsoft 365 Copilot profile and the Google Gemini profile. The eight-clause DPA hub structure lives at DPA for AI vendors.

1. Tenant model and data residency

| | Microsoft 365 Copilot | Google Workspace Gemini | |---|---|---| | Default location of customer content | Buyer's Microsoft 365 tenant | Buyer's Google Workspace tenant | | Default retention of AI interactions | Tenant-controlled (lives in the buyer's M365 substrate) | Set by the administrator: Gemini in Workspace prompts and responses from 90 days to indefinite; Gemini app conversations deleted after 3, 18 (the default) or 36 months, or kept for up to 72 hours when conversation history is off. Google keeps user feedback for up to 18 months | | Data residency commitment | EU Data Boundary service for tenants provisioned in the EU or EFTA (not Multi-Geo), with exceptions for flex routing at peak demand, Anthropic models and Bing web search queries | Workspace Data Regions (United States or Europe), covering Gemini prompts and responses but not Gemini Notebook. Google's Data Regions help page (updated 13 October 2026) gives storage at rest on most Business, Enterprise and Frontline editions (not Business Starter) and in-region processing on Enterprise Plus, Enterprise Essentials Plus and Frontline Plus or with the Enterprise Data Regions or Assured Controls add-ons. Since its version of 12 October 2026, the Workspace Service Specific Terms define the in-scope editions by reference to that page | | Tenant-bound default applies to | Microsoft-hosted Copilot features | Gemini in Workspace (formerly Gemini for Google Workspace) and the Gemini app used with a qualifying Workspace edition |

Both are tenant-bound by default. The buyer's compliance posture for the underlying M365 or Workspace tenant largely flows through to the AI feature, and the AI feature does not relocate the data, with an exception at Google (whose Data Regions page also excludes logs and cached content from Data Regions coverage). Gemini Notebook (formerly NotebookLM) creates a new copy of each file a user uploads from Drive and stores it with the user's Gemini Notebook data, and Google states that "Your organization's file sharing and data region settings do not apply to data in Gemini Notebook".

2. Training default on customer content

| | Microsoft 365 Copilot | Google Workspace Gemini | |---|---|---| | Use of customer content to train base models | No | No | | Use of prompts and responses to improve the AI feature | No, except optional user feedback, which Microsoft may use to improve Copilot (not to train foundation models) | No, except optional user feedback, which can include the prompt, contextual sources and output; Google may use it to improve its generative AI products, but not to train the models that support Workspace generative AI services | | Per-tenant fine-tuning available | Limited in M365 Copilot (Microsoft Copilot Tuning, in early access preview for a limited set of customers, tunes models with organisational data inside the Microsoft 365 tenant); broader for custom builds with Azure OpenAI models in Microsoft Foundry | None documented for Workspace; supervised fine-tuning is available on Gemini Enterprise Agent Platform (formerly Vertex AI) for custom builds | | Contract location | Microsoft Product Terms (Universal License Terms for Online Services, Use of Content for Training), not the DPA | Google Workspace Service Specific Terms, section 12.11 (Training Restriction), under the Google Cloud Terms of Service (which absorbed the Google Workspace Terms of Service in a version dated 1 June 2026, effective at renewal) or an offline Google Cloud Master Agreement |

The commercial default at both is protective. The Azure OpenAI fine-tuning route in Microsoft Foundry is customer-scoped: Microsoft states that "Fine-tuned models are exclusively available to the customer whose data was used to create the fine-tuned model", so the customer's training data does not enrich another tenant's model. At Google, fine-tuning is offered on Gemini Enterprise Agent Platform rather than in Workspace, and section 18 of the Google Cloud Service Specific Terms provides that Google will not use Customer Data to train or fine-tune any AI/ML models without the customer's prior permission or instruction.

3. Subprocessor depth and the Anthropic / OpenAI question

| | Microsoft 365 Copilot | Google Workspace Gemini | |---|---|---| | Underlying model provider | OpenAI GPT models (Azure OpenAI models operated by Microsoft and, since July 2026, OpenAI-operated models), Anthropic added 2026-01-07 | Google (Gemini family) | | Foundation model providers as subprocessors | Anthropic (since 7 January 2026) and OpenAI (since 23 June 2026, for OpenAI-operated models) are on the Microsoft Online Services Subprocessors List | Google models are first-party; no external foundation-model subprocessor | | EU Data Boundary scope of the Anthropic route | Explicitly out of scope as of 2026-01-07 | N/A | | Cloud subprocessor | Microsoft Azure (first-party) | Google Cloud (first-party) |

This is the cleanest point of divergence. Microsoft's Online Services Subprocessors List now brings Anthropic (since January 2026) and OpenAI, for OpenAI-operated models (since June 2026), into the buyer's data-flow diagram. Microsoft states that the Anthropic route is outside the EU Data Boundary. OpenAI-operated models are included in the boundary "except as otherwise noted in the EU Data Boundary documentation", which allows an AI subprocessor to process and store a pseudonymised user ID or session GUID outside the boundary, including in the United States. Anthropic is off by default for EU, EFTA and UK tenants, except that Copilot in Word, Excel and PowerPoint uses Anthropic by default in tenants there created after 25 March 2026. Flex routing, which lets Copilot inferencing leave the boundary at peak demand, is also on by default for eligible EU and EFTA tenants created after that date, and Bing web search, whose generated queries fall outside the boundary, is on by default in every commercial tenant. Buyers who chose Copilot in part because of the EU Data Boundary should check these settings and record whether each route is acceptable for the tenant.

Google Workspace Gemini avoids the question because the underlying models are Google's own, and the cloud is Google Cloud. Google's Workspace subprocessor list (last modified 23 July 2026) names 14 third-party companies, with their affiliates, engaged only for customer-initiated technical support or security risk detection, and none of them is a model provider.

4. EU/UK transfer mechanism

| | Microsoft 365 Copilot | Google Workspace Gemini | |---|---|---| | SCC version | 2021 SCCs implemented by Microsoft between Microsoft Ireland Operations Limited and Microsoft Corporation, processor-to-processor module (Module Three is our gloss), so the customer is not a party | The EU-U.S. Data Privacy Framework comes first (Google LLC certified; Google adopted it for EU data from 1 September 2023 and for UK and Swiss data from 16 September 2024; registry status Active, to 13 September 2027, when checked on 15 October 2026). Fallback: 2021 SCCs, Module 3 (processor to processor, Google exporter) between the Google contracting entity (Google Cloud EMEA Limited for most EMEA customers) and Google LLC, or Module 4 (processor to controller) with the customer where the Google entity is in an adequate country but the customer's billing address is not, and Module 2 or 3 with the customer where the Google entity is not in an adequate country | | UK Addendum | Yes | UK Extension to the Data Privacy Framework from 16 September 2024. In the fallback, the ICO's International Data Transfer Addendum is only in Google's controller-to-processor SCCs; the Module 3 (Google exporter) SCCs carry supplementary terms for Swiss transfers only | | EU Data Boundary commitment | Published, with named exceptions (flex routing at peak demand, Anthropic models, Bing web search queries) | Workspace Data Regions (see section 1); for Google Cloud workloads, the Assured Workloads EU Data Boundary control package restricts resources to EU regions | | Region-locked enterprise option | Microsoft Sovereign Cloud (formerly Microsoft Cloud for Sovereignty), with Sovereign Public Cloud, Sovereign Private Cloud and National Partner Cloud deployment models, for higher-assurance buyers | Assured Controls or Assured Controls Plus, paid add-ons to Workspace Frontline Plus or Enterprise Plus; Assured Workloads for Google Cloud workloads |

Both vendors publish the scope of their residency commitments: Microsoft in its EU Data Boundary documentation, and Google in its Workspace Data Regions pages, which list the editions, services and data states covered. For buyers in regulated industries (financial services, healthcare, public sector), the sovereign-cloud options at both vendors are the realistic answer rather than the standard Enterprise tier.

5. HIPAA BAA

| | Microsoft 365 Copilot | Google Workspace Gemini | |---|---|---| | BAA available | Yes. Microsoft's BAA is part of the DPA by default for covered entities and business associates (customers can opt out), Microsoft lists Microsoft Copilot and Copilot Chat in scope, and generated web search queries are excluded | Yes, once a Workspace super administrator accepts the BAA in the Admin console. Google lists Gemini in Google Workspace, the Gemini app (excluding Gemini in Google Chrome) and the Gemini Mac app among the covered Workspace services; Gemini Notebook is not covered | | Tier required | Any commercial or GCC customer that is a covered entity or business associate, for in-scope services | Any Google Workspace edition except the legacy free edition, once a super administrator accepts the BAA in the Admin console | | Consumer / SMB tier eligibility | Consumer Copilot is not in scope, and Microsoft's HIPAA page sets no business-size restriction | Consumer Gemini on personal Google accounts is not in scope. Business editions are eligible once an administrator accepts the BAA |

Google offers its BAA to Workspace customers on any edition except the legacy free edition, once an administrator accepts it, while Microsoft includes its BAA in the DPA by default for covered entities and business associates. Google's Workspace Service Specific Terms (section 12.7) also bar use of Workspace generative AI services for clinical purposes, medical advice, treatment or diagnosis, while leaving non-clinical research, scheduling and other administrative tasks unrestricted, and Gemini Notebook and Gemini in Chrome are outside Google's BAA. At either vendor, confirm that the services in use are in scope, have the BAA in place before any PHI enters the system, and run a fresh Security Rule risk analysis. See the HIPAA hub for the full checklist.

6. AI Act deployer obligations

Under the EU AI Act, the Gemini models are general-purpose AI models with Google as their provider, and Microsoft 365 Copilot runs on general-purpose AI models from OpenAI and Anthropic (see section 3). Each vendor supplies its product to the buyer as an AI system, so the deployer-side obligations fall on the buyer regardless of which vendor they pick. The use-case classification matters more than the vendor:

• A Copilot or Workspace Gemini deployment used for general office productivity (drafting emails, summarising meetings, finding documents) falls in minimal-risk or transparency territory.
• An agent used for recruitment or selection, or for promotion, termination or performance decisions, is high-risk under Article 6(2) and Annex III, point 4, from 2 December 2027 (a fixed date); the buyer has the Article 26 deployer obligations and should assess whether building or repurposing the agent makes it the provider (Articles 3(3), 25(1)(c)).
• An agent that evaluates the creditworthiness of individuals (not fraud detection) or prices life and health insurance is high-risk under Annex III, point 5, from the same date.

The EU AI Act hub covers the deployer assessment in full.

7. The Copilot Agents and Workspace integration angle

Copilot Studio and Google Workspace Studio (with AppSheet) both let the buyer build custom agents on top of tenant data. Section 13.3 of Google's Workspace Service Specific Terms (version of 12 October 2026) provides that "Customer is solely responsible for: (i) any Agentic AI Action", which the terms define to include any action or inaction of or attributable to an agent. This is where the AI Act exposure can climb fast:

• The agent's use case determines the AI Act risk tier, not the underlying platform.
• The custom data sources the agent connects to may expand the personal-data footprint beyond what the standard tenant DPA covers.
• Logging and human oversight controls become design-time buyer decisions, not vendor-default settings.

Buyers building agents on either platform should run the deployer assessment on the agent, not on Copilot or Gemini in general.

Picking between them

Pick Copilot when:

• The buyer is already on Microsoft 365 and the procurement gate values the published EU Data Boundary documentation
• The buyer needs the broader Microsoft Purview, Sensitivity Labels, and Conditional Access integration for DLP-driven controls on the AI surface
• The use case is HR-adjacent and the buyer wants Microsoft's Responsible AI Standard documentation as part of the deployer evidence base

Pick Workspace Gemini when:

• The buyer is already on Google Workspace and a fully first-party model stack is preferable (no Anthropic-style subprocessor question)
• The buyer wants an AI feature whose subprocessor list names no model provider, for simpler Transfer Impact Assessment documentation
• The buyer's developer workflows use Google Cloud and Gemini Enterprise Agent Platform (formerly Vertex AI) as the production AI surface and Workspace Gemini is the office-productivity layer above

Consider Microsoft Sovereign Cloud, Google Workspace Assured Controls or, for Google Cloud workloads, Assured Workloads when:

• The buyer is in financial services, public sector, or healthcare with sovereign-cloud requirements
• The standard EU Data Boundary or Workspace Data Regions is not enough on its own to clear the procurement gate

The agents question changes the answer

If the buyer's actual deployment plan is "build Copilot Agents" or "build Workspace integrations" rather than "use the off-the-shelf AI feature," the vendor selection question is downstream of the AI Act deployer assessment for the agent itself. Run that assessment first.

Related reading

• The eight DPA clauses both vendors are scored against: DPA for AI vendors
• The deployer-side AI Act read: EU AI Act for AI buyers
• The healthcare-specific procurement checklist: HIPAA for AI tools
• Cross-comparison: OpenAI vs Copilot enterprise compliance
• Cross-comparison: Gemini vs Vertex AI compliance

For ongoing AI compliance support, work with Janus DPO-as-a-Service. Browse the vendor index, other vendor comparisons, or the AI Agent Incident Register.