---
title: "Does Your AI Vendor Train on Your Data? Tier-by-Tier, 2026"
description: "A primary-source table of what ChatGPT, Claude, Microsoft Copilot and Google Gemini do with your content at every account tier: training defaults, EU data residency, retention and admin controls. Every claim carries the vendor doc it came from. Last verified 6 September 2026."
url: "https://ainora.lt/blog/does-your-ai-vendor-train-on-your-data"
published: "2026-09-05"
updated: "2026-09-06"
author: "Justas Butkus"
---

# Does Your AI Vendor Train on Your Data? Tier-by-Tier, 2026

**For the four assistants most European companies actually use, the answer splits on one line: consumer accounts may be trained on, organisational accounts are not.** A personal ChatGPT subscription, a personal Microsoft account in Copilot and a personal Google account in the Gemini app all sit on the training side of that line by default. ChatGPT Business and Enterprise, Claude Team and Enterprise, Copilot signed in with a work Entra ID, and Gemini inside a Google Workspace tenant all sit on the other side, and each vendor says so in its own documentation.

The nuance is where the money and the risk are. Not training is not the same as not storing. Not storing in the US is not the same as offering EU residency. And EU residency, where it exists, is never total, is sometimes impossible to apply retroactively, and in at least two documented cases does not extend to a product the same vendor sells you in the same subscription.

Every claim below comes from the vendor's own documentation, read on **6 September 2026**. Nothing here comes from a competitor summary, a secondary article, or a vendor's description of a rival. One qualification, because it matters for how much weight to put on a row: **OpenAI's help centre and marketing pages return HTTP 403 to automated retrieval**, so the OpenAI rows were read from archived copies of those same OpenAI pages rather than live. Every other vendor page here was read live.

## Does Your AI Vendor Train on Your Data?

“Trains on your data” is four separate questions that get collapsed into one, and collapsing them is what produces bad decisions. The four are: does the vendor use your content to improve its models; who inside the vendor can read it; where does it sit at rest; and how long does it stay there. A vendor can answer “no” to the first and still store your content indefinitely in a US data centre, which is exactly what one of the four does on its business tier.

This page answers all four, per product and per account type, with the vendor's own wording attached to each row. It is a reference sheet, not an argument. If a claim is not in a vendor doc, it is not here.

## Why Account Type, Not Price Tier, Decides It

The single most expensive misconception in a small or mid-sized company is that paying more buys privacy. It does not. A personal ChatGPT Plus subscription bought on a company card is still a consumer account, and OpenAI is clear that content from “services for individuals such as ChatGPT and Codex” may be used to train its models unless the individual opts out in the privacy portal. The same subscription bought as a ChatGPT Business seat is a different legal product with the opposite default.

Microsoft draws the same line by identity rather than by price. Its consumer FAQ lists who is excluded from model training, and the first entry is “Users signed into Copilot with an organizational Entra ID account.” The account you sign in with, not the licence you bought, decides which side of the line you are on. One consequence worth knowing: Microsoft also excludes “Users of Copilot within Microsoft 365 apps with Personal or Family subscriptions” from training, so a household subscription behaves differently from the free consumer Copilot on the same device.

> **A claim to be careful with**
>
> It is widely repeated online that Microsoft excludes the EEA from consumer Copilot training. That carve-out is **not** in Microsoft's own exclusion list, which names Brazil, China (excluding Hong Kong), Israel, Nigeria, South Korea and Vietnam. Microsoft's [privacy statement](https://www.microsoft.com/en-us/privacy/privacystatement) says only that “in some markets, this data can help train our AI models in Microsoft Copilot unless you opt out.” Do not plan around a European exemption that the vendor has not published. Check the toggle instead.
>
> [Source: Microsoft Support, Copilot privacy FAQ](https://support.microsoft.com/en-us/microsoft-copilot/privacy-faq-for-microsoft-copilot)

## Training Defaults, Tier by Tier

Read this by account type, not by brand. Each row carries the vendor document the wording came from.

| Product and account type | Trains on your content by default? | Who controls the setting |
| --- | --- | --- |
| ChatGPT Free, Plus or Pro (personal account)[Source: OpenAI Help Center, model training](https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance) | Yes, unless you opt out. “We may use your content to train our models.” | The individual, in the OpenAI privacy portal |
| ChatGPT Temporary Chat[Source: OpenAI Help Center, model training](https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance) | No. Temporary chats are “not used to train our models.” | The individual, per chat |
| ChatGPT Business (formerly Team)[Source: OpenAI Help Center, model training](https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance) | No. “By default, we do not train on any inputs or outputs from our products for business users.” | Workspace admin, opt-in only |
| ChatGPT Enterprise, Edu, Healthcare, for Teachers[Source: OpenAI, Enterprise privacy](https://openai.com/enterprise-privacy/) | No, “unless you have explicitly opted in to share your data with us.” | Workspace admin |
| OpenAI API Platform[Source: OpenAI Help Center, model training](https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance) | No. “Unless they explicitly opt-in, organizations are opted out of data-sharing by default.” | Organisation owner |
| Claude Free, Pro or Max (consumer)[Source: Anthropic Privacy Center, consumer](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training) | A setting, not a default anyone can quote. Anthropic lists training where “you choose to allow” it; the article does not state the shipped default. | The individual, in Privacy Settings |
| Claude Incognito chats[Source: Anthropic Privacy Center, consumer](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training) | No. “Not used to improve Claude, even if you have enabled Model Improvement.” | The individual, per chat |
| Claude Team, Enterprise, Anthropic API, Claude Gov[Source: Anthropic Privacy Center, commercial](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training) | No. “By default, we will not use your inputs or outputs from our commercial products … to train our models.” | Primary Owner or Owner |
| Microsoft Copilot, personal Microsoft account[Source: Microsoft Support, Copilot privacy FAQ](https://support.microsoft.com/en-us/microsoft-copilot/privacy-faq-for-microsoft-copilot) | Yes in some markets, unless you opt out. Training data includes “your voice and conversation activity with Copilot, including the images or files you upload.” | The individual, and the opt-out is partial |
| Microsoft Copilot, work or school account (Entra ID)[Source: Microsoft Learn, Copilot privacy](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy) | No. “Prompts, responses, and data accessed through Microsoft Graph aren’t used to train foundation LLMs.” | Tenant admin |
| Gemini app, consumer Google account[Source: Google, Gemini Apps Privacy Hub](https://support.google.com/gemini/answer/13594961) | Human review by default. “A subset of chats are reviewed by human reviewers … to help improve Google services.” | The individual, via Keep Activity |
| Gemini in Google Workspace[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | No. “Your content is not human reviewed or otherwise used for Generative AI model training outside your domain without permission.” | Workspace admin |

Two rows deserve a note. The **Claude consumer** row is deliberately not an answer. Anthropic's help article frames model improvement as something you “choose to allow,” and does not state whether the toggle ships on or off for a new account. Rather than repeat a guess, the honest instruction is: open Privacy Settings and look. Anthropic separately confirms that [commercial products](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training) are excluded by default, and that a Primary Owner or Owner can switch off the thumbs-up and thumbs-down button organisation-wide using the “Rate chats” setting under Organization settings, Data and Privacy.

A third row carries a qualifier worth reading slowly. Google's Workspace commitment is contractual and genuinely strong: prompts are “customer data under the Cloud Data Processing Addendum,” and Google undertakes “not to use customer data to train or fine-tune any of Google's generative artificial intelligence models supporting the Google Workspace Generative AI Services without our customer's prior permission or instruction.” The recurring qualifier in the shorter, more quotable version of that sentence is **“outside your domain”**. On its face that wording does not exclude use inside your own tenant, so the accurate summary is “not without permission, and not outside your domain,” not “Google never trains on your data.”

The **Gemini consumer** row is not phrased as training because Google does not phrase it that way. What Google publishes is human review: a subset of chats are read by reviewers, including trained service providers, to improve Google services. Google then gives the clearest single warning any of the four vendors offers, and it belongs on a slide in every induction deck: *“Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services.”*

## Where Does the Data Physically Live?

This is the section that changes procurement decisions, and it contains three facts that almost nobody publishes. One of them changed in March 2026 and has not yet reached most of the comparison content on the web, including some of ours.

> **Correction, 25 March 2026: Microsoft 365 Copilot may now process EU prompts outside the EU by default**
>
> Microsoft's Copilot privacy page still opens with the reassuring sentence, “EU traffic stays within the EU Data Boundary while worldwide traffic can be sent to the EU and other countries or regions for LLM processing.” Microsoft's own [EU Data Boundary exception register](https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-ongoing-partial-transfers) then records the exception: “To help maintain a consistent Copilot experience during periods of peak demand, Copilot prompts, responses, and grounding data may be processed outside the EU Data Boundary for AI inferencing, including in the United States, Canada, and Australia.” Microsoft calls the mechanism **flex routing**, and states that it is “on by default for eligible tenants that were created after March 25, 2026.” Tenants that existed before that date are told to check the Message Center for their own setting. The switch is real and it is yours: “EU and EFTA customers can disable flex routing in the Microsoft 365 admin center at any time,” and if you do, “LLM inferencing will occur inside the EU Data Boundary, even during periods of peak demand.” **Any page that says Copilot keeps EU prompts in the EU, full stop, is describing the product as it was before 25 March 2026.**
>
> [Source: Microsoft Learn, Copilot flex routing](https://learn.microsoft.com/en-us/microsoft-365/copilot/copilot-flex-routing)

**Second: Anthropic's first-party Claude does not store data in Europe.** Anthropic can route traffic to selected countries, including in Europe, but its privacy article ends the paragraph with six words that settle the question: “Note that data is stored in the US.” Its developer platform documentation is blunter still. Workspace geo, the setting that governs where data is stored at rest, “is set when you create a workspace and can't be changed afterward,” and “currently, `us` is the only available workspace geo.” Anthropic's own compliance page tells you where to go instead: “For Europe, you can select country-specific deployment options through AWS Bedrock, GCP Vertex, and Microsoft Foundry.” If a proposal tells you Claude is EU-hosted, ask which of those three it runs on.

**Third: Microsoft excludes Anthropic models from the EU Data Boundary, and this one is less alarming than it is usually reported.** Microsoft writes that “Anthropic models deployed in Microsoft offerings such as Microsoft Copilot, Researcher, Copilot Studio, Power Platform, and Copilot in Microsoft 365 apps are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments.” The sentence people leave off is the next one: “Customers within the EU Data Boundary and customers in the UK have Anthropic models disabled by default.” So this is an admin decision an EU tenant has to actively take, not something that happens to it. What is worth knowing is what the decision costs. If an admin enables the data-retaining variant, Microsoft says the contract stops covering it: “This means data is stored by Anthropic and not subject to your Microsoft Customer Agreement including commitments in the Product Terms and DPA,” and “Anthropic (not Microsoft) stores most inputs and outputs for up to 30 days before deleting them.”

A fourth item belongs with them, because it removes the boundary rather than punching a hole in it. Microsoft states that “customers who have purchased Multi-Geo Capabilities are not in scope for the EU Data Boundary even if their tenant is listed as being in a country or region in the EU or EFTA.” A multinational that bought Multi-Geo for a good reason may be outside a boundary its own policy documents assume it is inside.

| Vendor and product | EU data residency | The documented limit |
| --- | --- | --- |
| OpenAI, ChatGPT Enterprise and Edu[Source: OpenAI Help Center, data residency](https://help.openai.com/en/articles/9903489-data-residency-and-inference-residency-for-chatgpt) | Yes: storage at rest and GPU inference in Europe (EEA + Switzerland) | New workspaces only. Workspace metadata, names, billing, user logins and external integrations are listed as out of scope. |
| OpenAI, API Platform[Source: OpenAI, data residency in Europe](https://openai.com/index/introducing-data-residency-in-europe/) | Yes, per Project, with zero data retention on those requests | “European residency can only be configured for new Projects, existing Projects cannot be updated … after creation.” |
| OpenAI, consumer ChatGPT[Source: OpenAI Help Center, data residency](https://help.openai.com/en/articles/9903489-data-residency-and-inference-residency-for-chatgpt) | Not offered | Eligibility is limited to “eligible API customers and new ChatGPT Enterprise/Edu customers.” |
| Anthropic, first-party Claude (consumer, Team, Enterprise, API)[Source: Anthropic Privacy Center, server locations](https://privacy.claude.com/en/articles/7996890-where-are-your-servers-located-do-you-host-your-models-on-eu-servers) | Traffic routing only, not storage | “By default, we may route customer traffic to select countries in the US, Europe, Asia and Australia … Note that data is stored in the US.” |
| Anthropic, Claude Developer Platform workspaces[Source: Anthropic platform docs, data residency](https://platform.claude.com/docs/en/manage-claude/data-residency) | No EU option | “Workspace geo is set when you create a workspace and can’t be changed afterward. Currently, ‘us’ is the only available workspace geo.” |
| Anthropic, via cloud partners[Source: Anthropic, regional compliance](https://claude.com/regional-compliance) | Yes | “For Europe, you can select country-specific deployment options through AWS Bedrock, GCP Vertex, and Microsoft Foundry.” |
| Microsoft 365 Copilot, work or school account[Source: Microsoft Learn, Copilot privacy](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy) | Yes, with an exception that is on by default. “EU traffic stays within the EU Data Boundary while worldwide traffic can be sent to the EU and other countries or regions for LLM processing.” | Microsoft calls the exception flex routing and lists it in its own EU Data Boundary exception register. |
| Microsoft 365 Copilot, flex routing[Source: Microsoft Learn, EU Data Boundary ongoing partial transfers](https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-ongoing-partial-transfers) | Sends inferencing outside the boundary at peak | “Copilot prompts, responses, and grounding data may be processed outside the EU Data Boundary for AI inferencing, including in the United States, Canada, and Australia.” It is “on by default for eligible tenants that were created after March 25, 2026,” and EU and EFTA admins “can disable flex routing in the Microsoft 365 admin center at any time.” |
| Microsoft 365 Copilot, Anthropic models inside it[Source: Microsoft Learn, connect to an AI subprocessor](https://learn.microsoft.com/en-us/microsoft-365/copilot/connect-to-ai-subprocessor) | Outside the boundary, and off by default in the EU and UK | “Anthropic models deployed in Microsoft offerings … are currently excluded from the EU Data Boundary … Customers within the EU Data Boundary and customers in the UK have Anthropic models disabled by default.” |
| Microsoft 365 tenants with Multi-Geo[Source: Microsoft Learn, about the EU Data Boundary](https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-learn) | Out of scope entirely | “Customers who have purchased Multi-Geo Capabilities are not in scope for the EU Data Boundary even if their tenant is listed as being in a country or region in the EU or EFTA.” |
| Google Workspace with Gemini, storage[Source: Google Workspace, data covered by data regions](https://knowledge.workspace.google.com/admin/compliance/data-covered-by-data-regions) | Yes, from Business Standard upwards. “With data regions, you can choose to store your covered data in a specific geographic location (the United States or Europe).” | Business Starter has no data-regions control at all. Business Standard and Business Plus can set an at-rest region only. |
| Google Workspace with Gemini, processing[Source: Google Workspace, data region features by edition](https://knowledge.workspace.google.com/admin/compliance/compare-data-region-features-across-google-workspace-editions) | Yes, but gated to the top editions | “Data covered during processing varies by Google Workspace edition.” The processing-region policy sits with Enterprise Plus, Frontline Plus and the Data Regions add-on. On Business Standard and Business Plus the row for setting processing policies reads as unavailable. |
| Gemini Notebook, inside Workspace[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | The org policy does not reach it | “Your organization’s file sharing and data region settings do not apply to data in Gemini Notebook.” |

**Google is the one place where a widely repeated claim has flipped in the vendor's favour, and we are correcting it here rather than quietly leaving it.** It is still commonly written that Google Workspace data regions do not cover AI. That is now false. Google's data-regions table lists “Gemini App” and “Google Workspace with Gemini” with “Prompts and responses” as covered data, and ticks both the “Data covered at rest” and the “Data covered during processing” columns. What is true is that the coverage is **edition-gated**, and Google says so in a footnote on the same page: “Data covered during processing varies by Google Workspace edition.” Business Starter has no data-regions control at all; Business Standard and Business Plus can pin data at rest but not processing; the processing-region policy arrives with Enterprise Plus, Frontline Plus or the Data Regions add-on. Google is, on the published evidence, the only vendor in this comparison with a documented control over where inference happens. It is also the only one you have to buy the top edition to get.

Two more traps sit in that table. OpenAI's European residency **cannot be applied retroactively**: “European residency can only be configured for new Projects, existing Projects cannot be updated to have European data residency after creation,” and on the ChatGPT side the feature is offered to new Enterprise and Edu workspaces. A pilot that spins up projects in week one and asks about residency in week six has to be rebuilt, not reconfigured. And Google's own Workspace documentation records that **Gemini Notebook ignores the organisation's data-region and file-sharing settings**, which means the one Gemini surface most likely to be fed a folder of internal documents is the one your data-region policy does not reach.

Even where residency works, it is partial by design. OpenAI publishes what stays outside the chosen region: workspace metadata, workspace name, billing information, user logins, transient processing steps, and data handled through external integrations such as connected apps and web search. That is the useful teaching point. A residency badge is a statement about customer content, not about everything the service touches. The same instinct applies when you assess any vendor, which is why a written [vendor security assessment](/blog/ai-voice-agent-vendor-security-assessment-template) beats a sales call, and why [EU data residency claims](/blog/gdpr-compliant-voice-ai-platform-eu-data-residency-compared) are worth reading in the vendor's own words rather than in a comparison grid.

## How Long Is It Kept, and Does Deleting Delete?

Retention is where the defaults are least intuitive. The headline: **Claude for Work retains indefinitely unless an owner configures otherwise**, and the most-used consumer assistants keep 18 months by default on both Microsoft and Google.

| Product and tier | Retention default | What the doc says |
| --- | --- | --- |
| ChatGPT Business, Enterprise, Edu and Healthcare[Source: OpenAI, Enterprise privacy](https://openai.com/enterprise-privacy/) | Admin-set | Deleted conversations are “removed from our systems within 30 days.” |
| OpenAI API Platform[Source: OpenAI, Enterprise privacy](https://openai.com/enterprise-privacy/) | Up to 30 days | Zero data retention available “for eligible endpoints if you have a qualifying use-case.” |
| ChatGPT Temporary Chats[Source: OpenAI Help Center, Data Controls FAQ](https://help.openai.com/en/articles/7730893-data-controls-faq) | 30 days | “Temporary Chats are deleted from our systems after 30 days.” |
| Claude consumer, deleted chats[Source: Anthropic Privacy Center, consumer retention](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data) | Purged from back-end storage within 30 days | Removed from chat history immediately. |
| Claude consumer, model improvement enabled[Source: Anthropic Privacy Center, consumer retention](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data) | Up to 5 years, de-identified | Applies only to “new or resumed chats after the model training setting has been enabled.” |
| Claude, chats flagged by trust and safety[Source: Anthropic Privacy Center, consumer retention](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data) | 2 years for inputs and outputs | Classification scores are kept for up to 7 years. |
| Anthropic API[Source: Anthropic Privacy Center, org retention](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data) | 30 days | “We automatically delete inputs and outputs on our backend within 30 days,” with named exceptions. |
| Claude for Work (Team and Enterprise)[Source: Anthropic Privacy Center, custom retention](https://privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans) | Indefinite | “By default, data is retained indefinitely unless a custom retention period is set.” Minimum custom period: 30 days. |
| Microsoft Copilot, personal account[Source: Microsoft Support, Copilot privacy FAQ](https://support.microsoft.com/en-us/microsoft-copilot/privacy-faq-for-microsoft-copilot) | 18 months | “By default, we store conversation activity for 18 months.” |
| Microsoft 365 Copilot, work or school account[Source: Microsoft Learn, Copilot privacy](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy) | No default period published | Shared control, not admin-only. Admins “can also use Microsoft Purview to set retention policies for the data related to chat interactions with Copilot,” and users “can delete their Copilot activity history … by going to the My Account portal.” Microsoft publishes no default retention period for Copilot interaction data. |
| Microsoft 365 Copilot, what a user deletion proves[Source: Microsoft Learn, retention for Copilot and AI apps](https://learn.microsoft.com/en-us/purview/retention-policies-copilot) | Nothing, on its own | “Data from generative AI messages is stored in a hidden folder in the mailbox of the user who runs the AI app,” the clean-up “timer job typically takes 1-7 days to run,” and “messages visible in your AI apps are not an accurate reflection of whether they are retained or permanently deleted for compliance requirements.” |
| Gemini app, conversation history on[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | 18 months | “Admins can choose to automatically delete conversations after 3, 18, or 36 months. The default is 18 months.” |
| Gemini app, conversation history off[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | 72 hours, not zero | “When Gemini conversation history is off, new chats are saved in user accounts for up to 72 hours so Google can provide the service and process any user feedback.” |
| Gemini app, chats a human reviewer has seen[Source: Google, Gemini Apps Privacy Hub](https://support.google.com/gemini/answer/13594961) | Up to 3 years | “Not deleted when you delete your activity.” |
| Gemini in Google Workspace, the side panel[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | “90 days to indefinite, as determined by admins” | Indefinite is an option an admin can pick, not the shipped default. Admins “set automated retention periods based on inactive conversations (such as 3 months, 18 months, or 3 years), or retain conversations indefinitely for the organization,” and the user’s own deletion right, on by default, can be withdrawn by an admin. |
| Gemini Notebook, inside Workspace[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | Not retained after the session | “Gemini Notebook Prompts and responses Not retained after session ends None.” |

- **Indefinite** - Claude for Work default retention until an owner sets a custom period, minimum 30 days (Source: Anthropic Privacy Center, custom retention - https://privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans)
- **3 years** - How long a Gemini chat already seen by a human reviewer is kept, even after you delete your activity (Source: Google, Gemini Apps Privacy Hub - https://support.google.com/gemini/answer/13594961)
- **18 months** - Default conversation-activity retention on consumer Microsoft Copilot (Source: Microsoft Support, Copilot privacy FAQ - https://support.microsoft.com/en-us/microsoft-copilot/privacy-faq-for-microsoft-copilot)

The Gemini figure is the one to put in front of staff, because it answers the sentence people say after an incident: “but I deleted it.” Google records that human-reviewed chats and related data “are not deleted when you delete your activity.” Anthropic has an equivalent asymmetry on safety-flagged content, keeping inputs and outputs for two years and classification scores for seven, on both consumer and commercial products. Deletion controls are real, and they are not symmetrical with the vendor's own retention exceptions.

## What Can an Administrator Actually Set?

On business tiers, all four vendors give an administrator a genuine surface. The differences are less about how many switches exist and more about which risk each vendor thinks the switches are for.

| Vendor and tier | What an administrator can actually set |
| --- | --- |
| OpenAI, Business and Enterprise[Source: OpenAI, Enterprise privacy](https://openai.com/enterprise-privacy/) | How long data is retained, which apps and connectors are enabled for the workspace, enterprise authentication, and a Data Processing Addendum. Fine-tuned models are “for your use alone and never served to or shared with other customers.” |
| Anthropic, Team and Enterprise[Source: Anthropic Privacy Center, custom retention](https://privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans) | Custom data-retention periods, minimum 30 days, set by a Primary Owner or Owner. Changes are written to audit logs. |
| Microsoft, work or school Copilot[Source: Microsoft Learn, Copilot privacy](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy) | Purview retention policies for Copilot data, Content search and eDiscovery, sensitivity labels and rights management honoured, and: “Admins have full control to select which agents are allowed in their organization.” |
| Google Workspace[Source: Google Workspace, generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub) | Data-regions policy and Data Loss Prevention are applied automatically to Gemini, per-service enablement, and admin-set retention, with the documented Gemini Notebook exception. |

Microsoft's documentation contains the most useful sentence for a rollout plan, and it is not about the model at all: Copilot “only surfaces organizational data to which individual users have at least view permissions,” and Microsoft follows it with a warning that you must be using the permission models in SharePoint and elsewhere properly. Translated: for a Microsoft shop the real risk is not the assistant, it is the over-shared SharePoint the assistant can now search instantly. That is a permissions-hygiene project, not an AI project, and it is the highest-value thing to fix before a Copilot rollout.

Microsoft agrees, to the point of shipping a control for it and telling you it is temporary. [Restricted Content Discovery](https://learn.microsoft.com/en-us/sharepoint/restricted-content-discovery) “helps you limit discovery of content from specific SharePoint sites, including recently interacted files, in organization-wide search results and Microsoft Copilot responses while those reviews are taking place,” and Microsoft describes it as “a temporary governance control that gives organizations time to review and right-size access while continuing their Copilot deployment.” It also warns that leaning on it degrades the product you bought: “Excessive use can reduce the amount of content available to organization-wide search and Microsoft Copilot experiences, which can affect the completeness and relevance of search results and AI-generated responses.” And it is licence-gated, on top of Copilot itself: “Customers who are licensed for Copilot and have SharePoint Advanced Management available to them can configure Restricted Content Discovery.” We have written the failure mode up on its own, with the vendor documentation behind it, in [why an internal AI assistant surfaces documents people were never meant to find](/blog/internal-ai-assistant-permission-leakage).

## What Happens When Nobody Decides

The alternative to choosing is not neutrality. It is staff choosing for you, on personal accounts, on the training side of the line.

- **20%** - of studied organisations experienced breaches linked to shadow AI (Source: IBM, Cost of a Data Breach 2025 - https://www.ibm.com/think/insights/data-matters/cost-of-a-data-breach)
- **Up to USD 670K** - that shadow-AI incidents added to the average breach cost. IBM’s wording is “as much as USD 670K”, which is a ceiling and not a mean (Source: IBM, Cost of a Data Breach 2025 - https://www.ibm.com/think/insights/data-matters/cost-of-a-data-breach)
- **97%** - of organisations reporting AI-related breaches said they lacked proper access controls (Source: IBM, Cost of a Data Breach 2025 - https://www.ibm.com/think/insights/data-matters/cost-of-a-data-breach)
- **63%** - of breached organisations studied lacked AI governance policies (Source: IBM, Cost of a Data Breach 2025 - https://www.ibm.com/think/insights/data-matters/cost-of-a-data-breach)

Those four figures come from IBM's Cost of a Data Breach Report 2025. Read together they describe one failure, not four: the tools arrived before the decisions did. The 97% figure is the sharpest of them, because access control is precisely what an organisational account gives you and a personal account does not. The fix is not a ban. Bans push usage further into personal accounts, which is the expensive side of the table above. The fix is to sanction a tool, buy it on organisational identity, set the admin controls, and tell people which surface is for which kind of content.

## How to Decide What to Hand Your Staff

Four questions, in this order, answer it for most European companies.

**1. Which identity will people sign in with?** This decides the training default on every vendor in the table, so decide it first and enforce it with SSO. Everything downstream is easier once nobody is signing in with a personal account.

**2. Do you have a genuine EU storage requirement, or a preference?** If it is a requirement, the table narrows fast, and it narrows further if the requirement covers *processing* rather than only storage. OpenAI on new Enterprise or Edu workspaces will meet a storage requirement. Microsoft will, inside the EU Data Boundary, provided somebody has been into the admin center and turned flex routing off, has left Anthropic models disabled, and is not on Multi-Geo. Google Workspace will, and is the only one of the four with a documented control over the processing region as well, but that control arrives with Enterprise Plus rather than with Business Standard. Anthropic's first-party products will not meet it at all, and you will be looking at Bedrock, Vertex or Foundry. Decide before you provision, because on the OpenAI API side you cannot change it afterwards.

**3. What is the retention posture you want to be able to defend?** If you are in a sector where a regulator may one day ask how long prompts containing client data were kept, “indefinite by default” is not the answer you want to give, and neither is “the vendor does not publish a default.” Both of those appear in the table above, for different vendors. Set a custom period on day one, and on the Microsoft side that means somebody actually writing a Purview policy, because until they do there is no published period to point at.

**4. What are people allowed to paste?** This is the only one of the four that is a training question rather than a procurement question, and it is where an [AI literacy programme under Article 4 of the EU AI Act](/blog/eu-ai-act-article-4-ai-literacy) earns its place. The regulation does not require you to certify anyone or to guarantee a level. It does expect you to take measures appropriate to the systems you deploy and the people using them, and a documented rule about which surface takes which class of content is exactly that kind of measure.

The EU AI Act touches AI systems in other ways that are worth separating from data handling, because they are different duties with different deadlines. The Article 50 transparency obligations, for instance, govern whether a person must be told they are dealing with an AI system, which is a disclosure question rather than a data-residency one. We have covered those separately: [what the AI Act means for voice agents](/blog/eu-ai-act-voice-agents-what-businesses-need-to-know), [whether an AI caller must identify itself](/blog/eu-ai-act-transparency-requirements-ai-callers), [when a system counts as high-risk](/blog/eu-ai-act-high-risk-ai-classification-voice-systems), a practical [compliance checklist](/blog/eu-ai-act-voice-agent-compliance-checklist-2026), and the sector-specific case of [debt collection](/blog/eu-ai-act-debt-collection-compliance-2026). If you deploy someone else's system under your own brand, [who is responsible for what](/blog/is-white-label-ai-voice-gdpr-eu-ai-act-compliant) is the first thing to settle.

For the everyday side of this, our guide to [practical business uses of ChatGPT](/blog/chatgpt-for-business-10-practical-uses) covers what teams tend to do first, and the [GDPR guide](/blog/ai-voice-agent-gdpr-compliance-guide) and [security checklist](/blog/ai-voice-agent-security-compliance-checklist) cover the controls you should expect from any AI vendor you bring inside your processes.

## How This Page Is Kept Honest

> **Last verified 6 September 2026**
>
> Every vendor claim on this page comes from the vendor's own documentation, read on 6 September 2026, not taken from a summary. The OpenAI rows are the one qualification: those pages return HTTP 403 to automated retrieval, so they were read from archived copies of OpenAI's own pages rather than live, and should be re-checked in a browser before anything expensive rests on them. **Vendor terms change, sometimes without a changelog**, and several of the pages cited here were themselves updated within the previous month. This page is re-checked quarterly, and the next scheduled re-verification is December 2026. Where a doc has moved or a claim has changed, the row is corrected and the date at the top of this box moves with it.

**What the 6 September 2026 re-verification changed, so that you can see the cadence is real rather than advertised.** Four rows moved, all on the Microsoft and Google side.

- **Microsoft residency.** The page previously said Copilot is an EU Data Boundary service for EU customers and stopped there. It now carries flex routing, which has been on by default for eligible tenants created after 25 March 2026 and sends inferencing to the United States, Canada or Australia at peak. This was a change in the product, not an error in reading it, and it is the single most consequential item on the page.
- **Microsoft retention.** The row read “admin-set through Purview.” Microsoft's documentation describes a shared model in which admins set Purview policy and users can delete their own activity history, and it publishes no default period at all. Both halves are now stated, and the absence is stated as an absence.
- **Google residency.** The row asserted residency without the entitlement that governs it. Data-at-rest region policy starts at Business Standard; the processing-region policy is an Enterprise Plus, Frontline Plus or add-on entitlement. The old row was true for a tenant on the right edition and misleading for one on Business Standard.
- **Gemini app retention.** The row offered “indefinite” as one of the app's auto-delete options. Google's documentation enumerates three, 18 and 36 months for the Gemini app and names 18 months as the default; the indefinite option belongs to the Workspace side panel, where it is an admin choice rather than a default. The two have been separated.

Two cells in the tables above are deliberately not answers. The Claude consumer training default is a setting whose shipped state Anthropic's article does not publish, so the table says that rather than guessing. And the widely repeated EEA carve-out on consumer Copilot training is absent from Microsoft's own exclusion list, so it does not appear here at all. A third claim was considered and dropped: the frequently repeated line that **Microsoft Copilot retains interaction data indefinitely by default** could not be found on any Microsoft page, so this page publishes the absence rather than the claim. A table that is right about eleven rows and confidently wrong about one is worse than useless for the decision it is meant to support.

If you find a row that no longer matches the vendor doc it links to, the link is there so you can check it before we do.

## Frequently Asked Questions

**Does ChatGPT train on my data if I pay for Plus?**

Yes, unless you turn it off. Paying for a personal subscription does not change the account type. OpenAI states that for “services for individuals such as ChatGPT and Codex, we may use your content to train our models,” and points you to the privacy portal to opt out. The training default flips only when you move to ChatGPT Business, Enterprise, Edu or the API, where OpenAI does not train on inputs or outputs by default. (Source: OpenAI Help Center, model training - https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance)

**Is Claude data stored in the EU?**

Not on Anthropic’s first-party products. Anthropic can route traffic to selected countries including in Europe, but its own privacy article says plainly: “Note that data is stored in the US.” Its developer platform documentation is more explicit still, stating that “us” is currently the only available workspace geo and that the setting cannot be changed after the workspace is created. Anthropic points customers who need European deployment to AWS Bedrock, Google Cloud Vertex and Microsoft Foundry instead. (Source: Anthropic Privacy Center, server locations - https://privacy.claude.com/en/articles/7996890-where-are-your-servers-located-do-you-host-your-models-on-eu-servers)

**Is Microsoft Copilot covered by the EU Data Boundary?**

Partly, and less completely than it was before 25 March 2026. Microsoft still writes that “EU traffic stays within the EU Data Boundary while worldwide traffic can be sent to the EU and other countries or regions for LLM processing.” Its own EU Data Boundary exception register then records that “Copilot prompts, responses, and grounding data may be processed outside the EU Data Boundary for AI inferencing, including in the United States, Canada, and Australia.” Microsoft calls this flex routing and states it is “on by default for eligible tenants that were created after March 25, 2026.” EU and EFTA admins can switch it off in the Microsoft 365 admin center at any time, after which “LLM inferencing will occur inside the EU Data Boundary, even during periods of peak demand.” Two further carve-outs: Anthropic models inside Copilot are excluded from the boundary, though they are “disabled by default” for EU Data Boundary and UK customers; and tenants that bought Multi-Geo Capabilities “are not in scope for the EU Data Boundary” at all. (Source: Microsoft Learn, Copilot flex routing - https://learn.microsoft.com/en-us/microsoft-365/copilot/copilot-flex-routing)

**How long does Microsoft keep Copilot prompts by default?**

Microsoft does not publish a default. What it publishes is a mechanism: admins “can also use Microsoft Purview to set retention policies for the data related to chat interactions with Copilot,” and users “can delete their Copilot activity history, which includes their prompts and the responses Copilot returns, by going to the My Account portal.” That is a shared-control model, so a policy document that says only the admin controls it, or only the user controls it, is wrong either way. It is also worth knowing that a user deletion is not evidence of deletion: generative-AI messages sit “in a hidden folder in the mailbox of the user who runs the AI app,” the clean-up job “typically takes 1-7 days to run,” and Microsoft warns that “messages visible in your AI apps are not an accurate reflection of whether they are retained or permanently deleted for compliance requirements.” The claim circulating online that Copilot data is retained indefinitely by default is not supported by any Microsoft page we could find, so we publish the absence instead of the claim. (Source: Microsoft Learn, retention for Copilot and AI apps - https://learn.microsoft.com/en-us/purview/retention-policies-copilot)

**Do Google Workspace data regions cover Gemini prompts?**

Yes, and this corrects a claim still repeated widely. Google’s data-regions documentation lists “Gemini App” and “Google Workspace with Gemini” with “Prompts and responses” as covered data, ticked both for data covered at rest and for data covered during processing. The catch is the edition, not the data type. Google footnotes the same table with “Data covered during processing varies by Google Workspace edition”: Business Starter has no data-regions control, Business Standard and Business Plus can pin data at rest only, and the processing-region policy comes with Enterprise Plus, Frontline Plus or the Data Regions add-on. One documented exception survives regardless of edition: “Your organization’s file sharing and data region settings do not apply to data in Gemini Notebook.” (Source: Google Workspace, data covered by data regions - https://knowledge.workspace.google.com/admin/compliance/data-covered-by-data-regions)

**If I delete a Gemini chat, is it gone?**

Not necessarily. Google states that chats already seen by human reviewers, together with related data such as your language, device type and location information, “are not deleted when you delete your activity. Instead, they are retained for up to three years.” Deleting your activity removes the copy you can see, not the copy a reviewer already has. (Source: Google, Gemini Apps Privacy Hub - https://support.google.com/gemini/answer/13594961)

**Does opting out of training actually stop everything?**

No, and two vendors say so themselves. OpenAI notes that even if you have opted out, choosing to send feedback means “the entire conversation associated with that feedback may be used to train our models.” Microsoft states that its consumer opt-out “will not exclude your conversations from being used for other general product or system improvements nor from use for advertising, digital safety, security, and compliance purposes.” Treat the thumbs-up button as an export button. (Source: Microsoft Support, Copilot privacy controls - https://support.microsoft.com/en-us/microsoft-copilot/microsoft-copilot-privacy-controls)

**Can EU data residency be added to a workspace later?**

For OpenAI, not on the API side. The company states that “European residency can only be configured for new Projects, existing Projects cannot be updated to have European data residency after creation,” and its ChatGPT residency page limits eligibility to new Enterprise and Edu workspaces. A rollout that creates projects and workspaces first and thinks about residency second cannot be corrected without rebuilding them. (Source: OpenAI, data residency in Europe - https://openai.com/index/introducing-data-residency-in-europe/)

Related on AInora

Explore the platform and industry pages relevant to this article.

- [AINORA AI voice agentPlatform overview and capabilities](/ai-voice-agent)
- [AI debt collectionCompliant voice for recoveries](/ai-debt-collection)
- [PricingPriced individually, after a call](/pricing)
- [How it worksSetup, integrations, and go-live](/how-it-works)