AI Agents & Assistants
Microsoft 365 Copilot
Rest of world Report an error0–5 in half steps. 5 means the rubric's top anchor is met on the evidence.
by Microsoft Corporation · learn.microsoft.com
Compare with Gemini for Google Workspace → Report an error on this page Is this your product? →
Link to this page
Add ?theme=dark to the image address for dark backgrounds.
The badge shows the category and the month of the latest evaluation, and EU-Made where
that is the product’s provenance. It never shows a score.
Microsoft 365 Copilot at a glance
- Vendor
- Microsoft Corporation
- Headquarters
- Not stated on the vendor’s pages as of 26 Oct 2026
- Data hosting
- Not stated on the vendor’s pages as of 26 Oct 2026
- Free plan
- Not stated on the vendor’s pages as of 26 Oct 2026
- Panel verdict
- The AI panel rates Microsoft 365 Copilot 3/5, the mean of 6 judges. Microsoft 365 Copilot is an AI assistant add-on on top of Microsoft 365, and the bench scored it strongest on data use and training and on integrations and knowledge, both at 6-8: judges cited documented commitments that prompts, responses and Graph-accessed data are not used to train foundation models, a downloadable data protection addendum, ISO 27001 and ISO 42001 certifications, and a connectors API whose grounded answers store citations to their sources. how scores are made
- As of
- Vendor pages captured 26 Oct 2026; evaluated by the panel 28 Oct 2026. all sources
Read this page as one judge. Each weighs the same scores by what they care about.
The panel's verdict
Microsoft 365 Copilot is an AI assistant add-on on top of Microsoft 365, and the bench scored it strongest on data use and training and on integrations and knowledge, both at 6-8: judges cited documented commitments that prompts, responses and Graph-accessed data are not used to train foundation models, a downloadable data protection addendum, ISO 27001 and ISO 42001 certifications, and a connectors API whose grounded answers store citations to their sources. Sovereignty is where the judges genuinely split, spanning 4-6: the works council member credited the EU Data Boundary commitment in the Product Terms since March 2024 and a downloadable Data Protection Addendum, while the skeptic, head of IT and data protection officer weighed Anthropic's models being excluded from that boundary, the United States named among storage locations, and a US contracting vendor. Scores on processing and models span 5-6 on the same fault line, and agent capabilities 5-6, with documented jailbreak and prompt-injection classifiers but no public information on human approval before consequential actions. Pricing transparency scores sit at 2: no captured page states a price.
Speaks for it
- Documents that prompts, responses and Graph-accessed data are not used to train foundation models, backed by a downloadable data protection addendum and ISO 27001 and ISO 42001 certifications.
- Grounding surfaces only data a user has at least view permission for, with the Semantic Index honoring the identity-based access boundary.
- A documented connectors API with a five-step build, SDKs and Teams bots brings external data into Microsoft Graph, and grounded answers store citations to their sources.
- Users can delete their own Copilot activity history through the My Account portal, and admins manage stored prompts and responses through Purview and Content search.
- Guardrails are documented openly — jailbreak and cross-prompt injection attack classifiers, content harm filters, and protected-material detection.
Held against it
- The captured pages give different statements on training use: the Copilot privacy documentation says prompts, responses and Graph-accessed data are not used to train foundation models, while the general privacy statement says Microsoft "may use your data to develop and train our AI models".
- Anthropic's models are documented as currently excluded from the EU Data Boundary, and the privacy statement names the United States among storage locations.
- We found no public information on human approval before consequential agent actions, run logs, or triggers and schedules.
- We found no public information on stated retention periods or zero-data-retention terms with the named model providers, OpenAI and Anthropic.
- We found no public information on SSO with SCIM provisioning, group-level permissions per agent or connector, or exportable audit logs.
Best for
- You already run Microsoft 365 and want answers that respect each user's existing view permissions.
- You need to ground answers in external systems through a documented connectors API, with citations stored with each grounded response.
- Your procurement requires a downloadable data protection addendum and certifications such as ISO 27001 and ISO 42001.
- Your works-council review turns on pseudonymised system-generated logs and users' ability to delete their own Copilot activity history.
Avoid if
- You need every model behind the product to process inside the EU — Anthropic's models are documented as currently excluded from the EU Data Boundary.
- Your tenant uses Microsoft 365 Multi-Geo — ask the vendor: the public pages we read do not show it
- You are a small business unwilling to accept an annual commitment — the Microsoft Copilot Business add-on is documented with an annual commitment, up to 300 licensed users per tenant.
The scores
| Criterion | Panel median |
|---|---|
| Data use, training & retention | 3.5 of 5 |
| Models & processing location | 3 of 5 |
| Admin, identity & access controls | 3.5 of 5 |
| Integrations, connectors & knowledge | 3.5 of 5 |
| Agent capabilities & guardrails | 2.5 of 5 |
| European sovereignty | 2.5 of 5 |
Data use, training & retention
Show reasoningHide reasoning
How this is scored
What happens to prompts, uploaded files, connected data and outputs: whether any of it trains a model (the vendor's or a model provider's), how long it is kept and where logs live, whether zero-retention is available, and whether a data processing agreement covers it.
0 — No statement on training or retention; the privacy policy is a website policy that does not cover the product.
3 — A general statement that business data is "not used for training", with no word on retention, on the model providers behind the product or on logs.
5 — Business data excluded from training by contract, retention periods stated for chats and files, and a DPA offered to business customers.
8 — No-training commitments stated for the vendor and every model provider it routes to, retention configurable by the admin (including deletion of chat history), zero-data-retention terms with the model providers named, and abuse-monitoring retention described.
10 — Data use fully accounted for: training, retention, logging and abuse monitoring stated per model provider, customer-controlled retention and deletion with an audit trail, outputs and connected data treated as customer data in the DPA, and an independent audit (SOC 2, ISO 27001 or ISO 42001) covering the product.
The Head of IT
The Copilot privacy documentation states that prompts, responses and data accessed through Microsoft Graph aren't used to train foundation LLMs, that Copilot services have opted out of the human-review abuse monitoring available in Azure OpenAI, and that users can delete their own Copilot activity history, with a downloadable Data Protection Addendum and ISO 27001 and ISO 42001 certifications behind it. The captured pages do give different figures on training, though: the general privacy statement says Microsoft "may use your data to develop and train our AI models", while the product terms state that customer agreements take precedence over that statement. I found no public information on admin-configurable retention periods, stated retention durations for chats and files, or per-provider zero-data-retention terms. 3 5 6
The Data Protection Officer
Prompts, responses and Graph data are stated not to train foundation models, abuse monitoring with human review is opted out, a downloadable Data Protection Addendum applies, and users can delete their Copilot history via the My Account portal. But the privacy statement separately says Microsoft "may use your data to develop and train our AI models" — the captured pages give different statements on training use — and we found no public information on no-training or zero-data-retention commitments with the named model providers OpenAI and Anthropic, or on concrete retention periods the company controls. 3 4 5 6
The Automation Lead
Prompts, responses and Graph-accessed data are stated not to train the foundation models behind Copilot, a downloadable Data Protection Addendum covers business customers, users can delete their own Copilot history, abuse-monitoring human review is documented as opted out, and certifications include ISO 27001 and ISO 42001. The general privacy statement, though, says Microsoft "may use your data to develop and train our AI models" — enterprise agreements take precedence over it, but the captured pages give different statements on training, and I found no word on zero-data-retention terms with the named model providers or on admin-configurable retention of chat history. 3 4 5 6
The Customer-Service Director
The product documentation states prompts, responses and data reached through Microsoft Graph are not used to train the foundation models behind Copilot, describes abuse monitoring as opted out, shows users deleting their own Copilot history and admins managing stored interactions with Purview, and backs this with a downloadable data protection addendum plus ISO 27001 and ISO 42001 certifications. The general privacy statement, though, says Microsoft "may use your data to develop and train our AI models", so the captured pages give different statements on training, and I found no public information on zero-retention terms with the named model providers. 3 4 5 6
The Works Council Member
The product documentation states that prompts, responses and data reached through Microsoft Graph are not used to train foundation models, that Copilot services have opted out of abuse monitoring with human review, and that a downloadable Data Protection Addendum plus ISO 27001 and ISO 42001 certifications stand behind this. Stored prompts and responses are encrypted, personal data in system-generated logs must be pseudonymised, retention runs at the minimum time required per log type, users can delete their own Copilot history and admins manage stored data through Purview — pseudonymised logs are precisely what a works council needs. However, the general privacy statement says Microsoft may use your data to develop and train AI models, so the captured pages give different statements on training, and I found no public information on zero-retention terms with the named model providers. 3 4 5 6
The Skeptic
The Copilot privacy documentation commits that prompts, responses and Graph-accessed data are not used to train foundation models, is backed by a downloadable data protection addendum, and describes abuse monitoring with its human review opted out, plus ISO 27001 and ISO 42001 coverage. The captured pages give different figures on training, though: the general privacy statement says Microsoft "may use your data to develop and train our AI models". Retention is described only as minimum-time policies per log type with user-initiated deletion of activity history; I found no public information on stated retention periods, admin-controlled deletion, or zero-retention terms with OpenAI or Anthropic. 3 4 5 6
Models & processing location
Show reasoningHide reasoning
How this is scored
Which models sit behind the product and who runs them, where inference and storage happen, whether the customer can pin processing to the EU, choose or exclude model providers, bring its own model or key, or run the product in its own cloud or on premises.
0 — The models behind the product and the processing location are not stated.
3 — The model family is named, but where inference runs and through which provider (the model maker, a hyperscaler, the vendor) is unstated.
5 — Models and their providers named, with an EU processing option stated for storage and at least the default model.
8 — Each available model listed with provider and inference region, EU-only processing selectable by the admin and enforced across models, model providers excludable, and bring-your-own-key or own-cloud deployment documented.
10 — Processing fully under the customer's control and documented: EU-only or on-premises inference for every feature including agents and connectors, open-weight or self-hosted models supported, data flows per feature published, and a residency commitment in the contract rather than on a marketing page.
The Head of IT
OpenAI and Anthropic are named as the model subprocessors behind Copilot experiences, and the EU Data Boundary — with Copilot a covered workload in the Microsoft Product Terms since March 2024 — states that EU traffic stays inside the EU and EFTA for LLM processing. However, the same pages state Anthropic's models are "currently excluded from the EU Data Boundary", so EU processing is not enforced across every model behind the product. I found no public information on admin-selectable or excludable model providers, bring-your-own-key, or deployment outside Microsoft's cloud beyond the government community environments. 2 3 4
The Data Protection Officer
OpenAI and Anthropic are named as the model providers behind Copilot, and EU customers' traffic is stated to stay within the EU Data Boundary for storage and LLM processing, with Bing named for web search. But models provided by Anthropic are excluded from the EU Data Boundary, Multi-Geo customers are outside it entirely, and we found no public information on an admin-enforced EU-only option across all models, provider exclusion, bring-your-own-key or on-premises deployment. 3 4
The Automation Lead
OpenAI and Anthropic are named as the model subprocessors behind Copilot, and EU customers get the EU Data Boundary commitment with Copilot a covered workload since March 2024. Anthropic models are stated to be excluded from that boundary, worldwide traffic can be processed in the EU and other regions, and I found no documented way for an admin to pin every model to EU processing, exclude a provider, or bring their own key. 3 4 9
The Customer-Service Director
OpenAI and Anthropic are named as the model providers behind the product, and for EU tenants the EU Data Boundary keeps traffic inside the EU for language-model processing, a residency commitment carried in the Product Terms since March 2024. Anthropic's models are documented as excluded from that boundary, and I found no public information on excluding a provider, enforcing EU processing across every model, or bringing our own model or key. 3 4
The Works Council Member
OpenAI and Anthropic are named as model subprocessors, and EU tenants get storage and LLM processing inside the EU Data Boundary under commitments written into the Product Terms since March 2024. Against that, Anthropic models are excluded from the boundary, some data continues to be transferred outside it in limited circumstances, and models can be upgraded automatically — and I found no public information on an admin option to pin processing to the EU across all models, exclude a provider, or bring one's own key or cloud. 3 4 9
The Skeptic
OpenAI and Anthropic are named as the model providers, and EU tenants' Copilot traffic stays within the EU Data Boundary, a commitment written into the Product Terms since March 2024. But Anthropic's models are "currently excluded from the EU Data Boundary", so EU processing is not enforced across the model lineup. I found no public information on choosing or excluding a provider, bring-your-own-key, or on-premises deployment. 3 4
Admin, identity & access controls
Show reasoningHide reasoning
How this is scored
What the IT department controls: SSO and SCIM provisioning, roles and groups, which models, agents and connectors each group may use, retrieval that respects the permissions of the source system, audit logs, usage reporting, and policies such as data-loss prevention — and whether reporting on staff use is scoped and documented.
0 — No admin features documented; accounts are individual sign-ups.
3 — A team workspace with an owner and member invitations, but no SSO, no roles and no audit log.
5 — SSO, admin and member roles, user management, and an admin view of usage.
8 — SAML/OIDC SSO with SCIM provisioning, group-based permissions for models, agents and connectors, retrieval that inherits source-system permissions, exportable audit logs, and usage analytics with documented scope.
10 — Governance a regulated company can operate: fine-grained policies (DLP, content filters, prompt and output logging rules), audit logs streamed to a SIEM, per-agent ownership and approval, usage reporting that can be aggregated or pseudonymised for works-council agreements, and every admin action logged.
The Head of IT
This is documented well where it matters most for my rollout: Copilot only surfaces organizational data a user has at least view permission for, the Semantic Index enforces the identity-based access boundary during grounding, Purview-protected content keeps its usage rights, and admins can select which agents are allowed in the organisation — our SharePoint permission model carries through to answers. Entra authorization with role-based access control, Content search and Purview for stored prompts and responses, Teams export APIs for chat data, and PowerShell-based usage reports give me real admin tooling. I found no public information on SCIM provisioning, group-level assignment of specific models, or streaming audit logs to a SIEM. 3 4 7
The Data Protection Officer
Retrieval inherits source-system permissions — Copilot only surfaces data the user has view rights to and the Semantic Index honors the identity-based access boundary — alongside tenant isolation through Entra authorization and role-based access control, admin control over which agents are allowed, Purview information-protection labels honored, and PowerShell-based user, group and usage reporting. We found no public information on SSO with SCIM provisioning, group-scoped permissions for models and connectors, or exportable audit logs on the captured pages. 3 7
The Automation Lead
Retrieval inherits source-system permissions twice over — Copilot surfaces only data the user has at least view rights to and the Semantic Index honours the identity boundary — and admins can select which agents are allowed at all, with Purview and Content Search to manage stored interactions and Entra plus role-based access control for tenant isolation. What holds it back: the captured pages do not document SCIM provisioning, exportable audit logs, or group-level permissions per model and connector, and the scope of usage reporting is not spelled out. 2 3 7
The Customer-Service Director
Retrieval provably respects source permissions — Copilot only surfaces data a user has at least view rights to, the semantic index honors the identity boundary, and Purview sensitivity labels are honored — and admins can block agents and switch off connected experiences, with tenant isolation through Entra and role-based access control. I found no public information on SCIM provisioning, group-level permissions per agent or connector, or exportable audit logs of Copilot use. 3 4 7
The Works Council Member
Retrieval honours the individual user's view permissions in both the Semantic Index and grounding, tenants are isolated through Entra authorization and role-based access control, and admins decide which agents and connected experiences staff may use, with Microsoft 365 usage reports available through PowerShell. I found no public information on single sign-on with SCIM provisioning stated in so many words, group-level permissions per agent or connector, exportable audit logs, or usage reporting documented as aggregatable or pseudonymised for a works-council agreement — the last point is where a co-determination conversation would start. 3 4 7
The Skeptic
Grounding only surfaces content a user has view permissions for and the Semantic Index honors the identity-based access boundary — genuine, documented permission inheritance — and admins can decide which agents are allowed and switch Copilot off per app, with Purview and Content search to inspect stored prompts and responses. I found no public information on SSO or SCIM provisioning, group-level control over models and connectors, or exportable audit logs and usage analytics. 3 4 7
Integrations, connectors & knowledge
Show reasoningHide reasoning
How this is scored
How the product reaches company knowledge and systems: connectors to file stores, wikis, mail, CRM, ticketing and ERP, how fast and how completely they sync, the API, support for open protocols such as MCP, and how the assistant cites the sources it answers from.
0 — No integrations; users paste text or upload files one at a time.
3 — File upload and one or two connectors (for example a drive), with no API and no statement on sync or citations.
5 — A connector library covering the common file stores, wikis and collaboration tools, answers citing sources, and a documented API.
8 — Connectors to file stores, wikis, mail, CRM and ticketing with incremental sync, custom connectors or MCP support, an API and webhooks for embedding assistants and agents in other tools, and source citations on every grounded answer.
10 — The product sits in the company's systems: read and write connectors with documented scopes, MCP servers and clients, contact-centre or telephony integration where relevant, a knowledge layer that can be audited and corrected, and the data, agents and configuration exportable when the customer leaves.
The Head of IT
Copilot connectors bring external data into Microsoft Graph through a documented five-step build using the connectors API, with per-item content access settings and natural-language search over the ingested content, and stored responses include citations to the sources that grounded them. Graph REST APIs, SDKs, change notifications with Azure Event Hubs, Teams bots and the Copilot APIs let us embed assistants and agents into other tools. I found no public information on a named catalogue of connectors for CRM or ticketing systems, incremental sync frequency, or MCP support. 3 7 8 9
The Data Protection Officer
A documented connectors API with a five-step build process brings external data into Microsoft Graph, Graph REST APIs and SDKs cover mail, Teams bots and change notifications, and stored interactions include citations to the sources used to ground responses. We found no public information on the catalog of ready-made connectors, sync frequency and completeness, or support for open protocols such as MCP. 3 7 8
The Automation Lead
There is a real connector pipeline — bring external data into Microsoft Graph, register schemas with semantic labels, manage access to external items — plus a documented connectors API, Graph REST APIs with SDKs, change notifications and Teams bots, and grounded answers store citations to what they used. I found no public information on MCP support or incremental sync behaviour on the captured pages, which is what an automation shop would probe next. 3 7 8
The Customer-Service Director
There is a real connector framework — connectors that pull external data into Microsoft Graph for grounding, a documented connectors API with a five-step build, Graph REST APIs, SDKs, change notifications and Teams bots — and every grounded answer carries citations that are stored with the interaction. No CRM or ticketing connector is named on the captured pages, sync frequency is not stated, and I found no public information on MCP support. 3 7 8
The Works Council Member
The Copilot connectors framework brings external systems into Microsoft Graph through a documented five-step build, a connectors API, SDKs, code samples and a quick start, with change notifications and Teams bots to embed the assistant in other tools. Grounded answers carry citations, and the stored data explicitly includes citations to the sources used, so staff can see where an answer came from. 3 7 8 9
The Skeptic
There is a real connector story: a documented Copilot connectors API with build steps, semantic schema labels and external item ingestion, natural-language search over ingested content, plus Graph REST APIs, SDKs, a mail API, Teams bots and change notifications for embedding elsewhere. Grounded responses are stored with citations to their sources. I found no public information on connector sync frequency or support for open protocols such as MCP. 3 7 8 9
Agent capabilities & guardrails
Show reasoningHide reasoning
How this is scored
What an agent can actually do and where it stops: multi-step tasks, tool use and actions in other systems, scheduling and triggers, human approval before consequential actions, run logs, evaluation and testing before release, and the limits the vendor states rather than hides.
0 — No agents; the product is a chat window, whatever the marketing says.
3 — Saved prompts or custom assistants with instructions and files, but no actions in other systems and no run history.
5 — Agents that call tools or APIs and run multi-step tasks, built without code, with a log of each run.
8 — Agents that act in connected systems with scoped permissions, human approval configurable per action, triggers and schedules, test and evaluation tooling before release, and documented limits and failure handling.
10 — Agents a business can put in production and account for: versioning and staged rollout, evaluation against test sets, monitoring of quality and cost per agent, handover to a human with full context, every action attributable to the agent and the person who deployed it, and the vendor's own statement of what agents must not be trusted with.
The Head of IT
Copilot Studio builds agents without code on a visual canvas, the Agents Toolkit runs them across Copilot, Teams, web and third-party messaging channels, and the developer flow explicitly covers creating, testing and deploying agents, with admins able to select which agents are allowed in the organisation. Guardrails are documented openly — jailbreak and cross-prompt injection attack classifiers, content harm filters across hate and fairness, sexual, violence and self-harm categories, and protected-material detection for copyrighted text and licensed code. I found no public information on human approval before consequential actions, per-run logs, triggers or schedules, or evaluation against test sets before release. 3 9
The Data Protection Officer
Agents are built without code on the Copilot Studio visual canvas or in Visual Studio Code, deploy across Copilot, Teams, web and third-party channels, and ship with jailbreak and cross-prompt injection classifiers, content harm filters, protected-material detection and admin control over which agents are allowed. We found no public information on human approval before consequential actions, run logs, evaluation against test sets before release, or a vendor statement of what agents must not be trusted with. 3 9
The Automation Lead
Copilot Studio builds agents on a visual canvas with pre-built tools, the Agents Toolkit deploys across Copilot, Teams, web and third-party messaging channels, and agents can be created, tested and deployed from Visual Studio Code — so there is a test-before-release story, plus jailbreak and cross-prompt-injection classifiers and admin gating of which agents may run. What I could not find on the captured pages: per-action human approval, run logs, triggers and schedules, or scoped per-agent permissions — and for agents that file tickets and chase approvals, those guardrails have to be in writing. 3 7 9
The Customer-Service Director
Copilot Studio builds agents on a visual canvas with pre-built tools that deploy across Copilot, Teams, web and third-party messaging channels, agents can be created, tested and deployed from developer tooling, admins control which agents are allowed, and guardrails such as jailbreak and prompt-injection classifiers, harm filters and protected-material detection are documented. What I cannot find is what my contact centre lives or dies on: human approval before consequential actions, run logs, triggers and schedules, evaluation against test sets before go-live, and handover to a human agent with the full conversation context. 3 7 9
The Works Council Member
Agents are built without code on a visual canvas in Copilot Studio and the Agents Toolkit, run across Copilot, Teams and third-party messaging channels, and admins control which agents are allowed, with jailbreak and cross-prompt-injection classifiers and content filters as guardrails. But I found no public information on human approval before consequential actions, a log of each run, triggers or schedules, or the vendor's own statement of what agents must not be trusted with — a human approving what an agent does to a colleague is my condition for signing off, and nothing captured addresses it. 3 9
The Skeptic
Copilot Studio builds agents with pre-built tools on a visual canvas, the Agents Toolkit deploys them across Copilot, Teams, web and third-party channels, admins gate which agents are allowed, and jailbreak and cross-prompt-injection classifiers plus content harm filters are documented. Creating and testing agents is described, but I found no public information on run logs, human approval before consequential actions, triggers and schedules, or a stated limit on what agents should not be trusted with — I would not run these unsupervised on this evidence. 3 9
European sovereignty
panel opinion
Show reasoningHide reasoning
How this is scored
Who the contracting entity and the parent company are, where customer data is hosted, and who the subprocessors are — model providers above all, then hosting, search, telephony and email. Independently sourced by the sovereignty pipeline; for an AI product the model provider is a subprocessor, so a European interface over a non-European model counts as such.
0 — Non-EU vendor and contracting entity, hosting unstated, subprocessors (including model providers) unnamed.
3 — A privacy policy exists, but hosting location is unstated or non-EU and the model providers behind the product are not named as subprocessors.
5 — EU hosting stated and an EU contracting entity, but prompts reach non-EU model providers or other subprocessors without a stated safeguard or EU region.
8 — EU hosting on named infrastructure, EU contracting entity, every subprocessor listed with its location including model providers, EU inference available, and a DPA offered.
10 — Sovereign end to end and evidenced: European ownership, entity, hosting and model processing, every subprocessor European or optional, data export and deletion documented, and a security or certification statement published rather than implied.
The Head of IT
Microsoft Corporation is a US company in Redmond and I found no public information on an EU contracting entity, so European sovereignty rests entirely on the EU Data Boundary — a contractual commitment that EU tenants' customer data is stored and processed within the EU and EFTA. Inside that boundary OpenAI and Anthropic are named as model subprocessors, but the captured pages state Anthropic's models are excluded from the boundary and that data continues to be transferred outside it in limited circumstances. The Data Protection Addendum and EU-U.S. Data Privacy Framework membership are published safeguards, but the model chain is not European end to end. 3 4 5 6
The Data Protection Officer
The EU Data Boundary commits to storing and processing Microsoft 365 customer data inside the EU and EFTA, OpenAI and Anthropic are named as model subprocessors, and a Data Protection Addendum is downloadable — but Anthropic's models are excluded from that boundary, Multi-Geo customers are outside it entirely, and transfers out continue in "limited circumstances". The contracting entity is a US company and we found no public information on a European contracting entity or European model processing for every feature. 3 4 5
The Automation Lead
The EU Data Boundary is a contractual commitment in the Product Terms with Copilot a covered workload, a Data Protection Addendum is offered, and the model providers — OpenAI and Anthropic — are named as subprocessors. The boundary is not total, though: Anthropic models are excluded from it, some customer and professional-services data continues outside in limited circumstances, the privacy statement says personal data may be stored in the United States and other jurisdictions, and the vendor here is Microsoft Corporation, Redmond, USA — so what is proven is EU hosting and named subprocessors, not a European chain end to end. 3 4 5 6
The Customer-Service Director
For EU tenants the EU Data Boundary commits to storing and processing customer data inside the EU, a commitment in the Product Terms since March 2024, with a downloadable data protection addendum and Data Privacy Framework safeguards. But the vendor is Microsoft Corporation of Redmond, Anthropic model processing is documented as outside the boundary, some data continues to transfer out in limited circumstances, and the United States is named among storage locations. 3 4 5 6
The Works Council Member
EU storage and processing inside the EU Data Boundary is a Product Terms commitment covering Microsoft 365 Copilot on Azure regions, with OpenAI and Anthropic named as model subprocessors and a downloadable Data Protection Addendum offered. Against that stand the US parent in Redmond, a contracting entity whose jurisdiction the captured pages do not state, the exclusion of Anthropic models from the boundary, and continuing transfers outside it in limited circumstances. 3 4 5
The Skeptic
The contracting vendor is Microsoft Corporation of Redmond, and the EU Data Boundary keeps EU tenants' Copilot traffic in the EU — yet Anthropic's models are excluded from that boundary and the privacy statement names the United States and other jurisdictions among storage locations. OpenAI and Anthropic are named as subprocessors and a downloadable DPA exists, but the stated cross-border safeguard is the EU-U.S. Data Privacy Framework rather than European processing. I found no public information on an EU contracting entity or on the locations of the subprocessors. 3 4 5 6
Pricing transparency
not rated — the vendor publishes no price
Show reasoningHide reasoning
How this is scored
Whether a company can compute the real annual cost from public pages alone: price per seat, per agent, per conversation or per usage unit, what is included in each tier, usage limits and overage, which models, connectors or agents cost extra, minimum seats and term.
0 — No public prices at all; every customer is a sales call.
3 — A starting price or "from" figure exists, but usage limits, included models and what scales the cost are unclear.
5 — Prices per tier are public with the main inclusions, but usage limits, overage, agent runs or enterprise features (SSO, audit logs, EU processing) are unpriced or "on request".
8 — Every tier priced publicly with seat or usage units, usage limits and overage rates, add-ons listed, minimum seats and term stated.
10 — Complete price computability: annual cost derivable for a given number of seats, agents or conversations, with every usage rate, model surcharge, add-on and support tier and the cancellation and data-return terms published.
The Head of IT
The licensing pages document the eligible plans (Office 365 E1 through E5 and F3), a Business add-on capped at 300 users with an annual commitment, and the education and government purchase channels — but they publish no price per seat, so I cannot compute what a rollout to eight hundred staff would cost. I found no public information on usage limits, add-on pricing, or what enterprise features cost. For a purchase of this size, the number comes from a negotiation, not from a page. 2 7
The Data Protection Officer
The licensing structure is public — an add-on to eligible Office 365 E5, E3, E1 and F3 plans, an annual-commitment Business option capped at 300 users, plus education, government and small-business channels. But we found no public price figures on the captured licensing pages, so a company cannot compute even a starting annual cost from them. 2
The Automation Lead
The licensing pages document real structure — eligible Office 365 plans, an annual-commitment Business add-on capped at 300 users per tenant, education and government channels — but I found no public price figure on the captured pages: no per-seat amount, no usage limits, no overage rates. A buyer cannot compute an annual cost from these pages alone. 2
The Customer-Service Director
The licensing pages document structure in detail — eligible Office 365 plans, an SMB add-on with an annual commitment capped at 300 users, education purchases through EES or CSP, and government cloud availability — but I found no public information on actual prices: no per-seat figure, let alone the per-conversation price my CFO needs to forecast the annual cost. 2 7
The Works Council Member
The licensing pages name the eligible base plans, an add-on for organisations up to 300 users with an annual commitment, and education and government purchasing channels, but no captured page states a price per seat, usage limits, or what scales the cost. A company cannot compute its annual spend from these pages alone. 2 7
The Skeptic
The licensing pages document which plans are eligible (Office 365 E1 through E5 and F3), that the SMB add-on requires an annual commitment and supports up to 300 users per tenant, and that education is bought through EES or CSP — but no price figure appears anywhere I read. I found no public information on a price per seat, usage limits, or enterprise terms; the cost of a thousand seats is not computable from these pages. 2 7
European sovereignty — proven facts
0 of 4 dimensions provenBuilt only from facts shown on the vendor's own pages. A dimension we could not prove is left open, not scored as zero.
| Legal entity | Not determined | — | uncited Report an error |
|---|---|---|---|
| Ownership | Not determined | — | uncited Report an error |
| Data residency | Not determined ⚠ unverified | — | uncited Report an error |
| Subprocessors | Not determined | — | uncited Report an error |
Questions about Microsoft 365 Copilot
Is Microsoft 365 Copilot European?
No. whats-best.ai classifies Microsoft 365 Copilot as Rest of world.
Where does Microsoft 365 Copilot store data?
Not stated on the vendor’s pages as of 26 Oct 2026.
Does Microsoft 365 Copilot have a free plan?
Not stated on the vendor’s pages as of 26 Oct 2026.
What are the best alternatives to Microsoft 365 Copilot?
As of 28 Oct 2026, the panel rates Langdock (4/5), Dust (3.5/5) and Unique AI (3.5/5) highest among the other 15 AI Agents & Assistants products evaluated. Ratings that look equal are rounded to half stars; the order uses the unrounded rating, then the name. The highest-rated European alternative is Langdock (EU-Made). Ratings are the AI panel’s opinion, formed from public evidence: how scores are made. The full list, with European products first: Microsoft 365 Copilot alternatives.
Where this could be wrong
- Evidence ages. The oldest capture behind this page is from 26 Oct 2026. Vendors change pricing and policies without notice; every fact reflects its source as of the capture date shown in the registry.
- Weak sourcing — Data residency. Not confirmed on the vendor’s own pages as captured.
- AI can misread a source. Extraction and judgement are automated; a citation guarantees traceability, not infallibility. If something here is wrong, say so — no account needed, every report is decided within 5 business days, and accepted corrections are published.
What we left out
A claim that does not survive our checks costs us the claim, not the page. This is what was taken off this one.
- 4 pricing facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 3 compliance facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 product facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 1 hosting fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 sovereignty dimension could not be confirmed on the vendor’s own pages and is shown as unknown. Know more? Tell us
Sources (9)
The pages every claim on this page was read from — each one checked, dated, and kept verifiable.
- 1 Product documentation (overview) learn.microsoft.com Checked 26 Oct 2026 Details →
- 2 Licensing and pricing documentation learn.microsoft.com Checked 26 Oct 2026 Details →
- 3 Data, privacy and security documentation learn.microsoft.com Checked 26 Oct 2026 Details →
- 4 EU Data Boundary documentation learn.microsoft.com Checked 26 Oct 2026 Details →
- 5 Data protection addendum www.microsoft.com Checked 26 Oct 2026 Details →
- 6 Privacy policy www.microsoft.com Checked 26 Oct 2026 Details →
- 7 Integrations, connectors & knowledge — found from sitemap developer.microsoft.com Checked 26 Oct 2026 Details →
- 8 Integrations, connectors & knowledge — found from sitemap developer.microsoft.com Checked 26 Oct 2026 Details →
- 9 Agent capabilities & guardrails — found from sitemap developer.microsoft.com Checked 26 Oct 2026 Details →