Yes - if your permissions are right. Microsoft Copilot and Google Gemini answer only from files the asking user can already open; the AI inherits your existing access rights. The real risk is not the AI leaking across accounts, but years of sloppy sharing - "anyone with the link" folders - suddenly becoming easy to find. Audit sharing before you switch it on.
How access actually works
Both vendors state the same principle: the AI respects your existing permissions. When an employee asks a question, the system retrieves only from content that person is already allowed to see - and answers include links to the source documents. Someone without access to the salary folder does not get answers from it.
The real risk: your old sharing habits
Over-shared files were mostly invisible because nobody went looking. An AI finds them with one question. Typical surprises:
Folders shared with "anyone in the organisation with the link" from years ago.
External guests still on project folders from finished engagements.
HR or finance documents sitting in broadly shared drives.
This is a permissions hygiene problem the AI makes visible - fixable, and worth fixing regardless of AI.
What the vendors promise about your data
Microsoft states that prompts and responses in Microsoft 365 Copilot are not used to train the underlying foundation models, and that Copilot surfaces only content the user already has access to.
Google states that for business Workspace tiers, your content is not used to train Gemini models without your permission, and Gemini respects your existing Drive sharing settings.
Read the current terms for your specific licence before relying on this - business and consumer tiers differ.
A pre-flight checklist
1. Audit link-sharing and guest access on the drives in scope.
2. Classify folders: which are fine for broad AI search, which are not.
3. Switch on for a pilot group first and review the audit logs.
4. Tell employees what the AI can and cannot see - and how to report a surprise.
When a custom setup makes sense
Regulated data (banking secrecy, patient data), strict Swiss data residency requirements, or sources spanning several systems are good reasons for a custom retrieval setup where you choose the hosting and the model - rather than relying on a suite's built-in assistant.
Sources
Microsoft, "Data, Privacy, and Security for Microsoft Copilot" (accessed 20 October 2026)
Google, "Generative AI in Google Workspace Privacy Hub" (accessed 20 October 2026)
Notion, "Notion AI security & privacy practices" (accessed 20 October 2026)
Last updated: 20 October 2026.
Want an AI that answers from your files without exposing them? Discuss a project ->
