Concerns like this can be addressed in a few concrete ways:
- Review and manage what Copilot “remembers”
Copilot can personalize responses by remembering key details from past conversations (called Memory or personalization). It may summarize or restate information in ways that feel unfamiliar or incorrect. To see and correct this:- Ask Copilot: “What do you know about me?” to see what it currently remembers.
- If something is wrong, ask it to forget that item, for example: “Forget that I asked about X”.
- To delete everything from memory, select the profile icon in Copilot, then Memory > Delete all Memory.
- If desired, turn off personalization entirely using the Copilot privacy controls so Copilot no longer remembers details from conversations.
- Turn off or limit personalization based on chat history
If using Microsoft 365 Copilot with a work or school account:- Go to Copilot Chat and open Settings → Personalization.
- In Chat history, turn the toggle Off. This stops Copilot from using previous chats to personalize responses and schedules existing inferences for deletion.
- Delete Copilot interaction history
If there is concern about past interactions being stored:- Go to
https://myaccount.microsoft.com→ Settings & Privacy → Privacy → Copilot interaction history. - Use the option to clear Copilot history. This removes Copilot interaction data across Microsoft 365 apps.
- Go to
- Control what Copilot remembers and how it uses data
Copilot provides privacy controls to:- See, change, or delete what it remembers.
- Turn off personalization so Copilot no longer keeps memories of conversations.
These options are available when signed in with a Microsoft account or supported work/school account.
- Provide feedback to Microsoft
Microsoft is interested in feedback about incorrect, confusing, or concerning behavior:- In the Copilot experience, use the built-in feedback mechanism (thumbs down / feedback form) on the specific response that misrepresented what was said.
- When submitting feedback, Microsoft may collect the generated response, relevant content samples (such as chat history used to produce the answer), and diagnostic logs to understand what happened. This data is only sent when feedback is submitted.
- Level of concern
Generative AI systems can produce inaccurate or misleading content (“hallucinations”), including incorrect summaries of prior conversations. Microsoft explicitly advises reviewing, editing, and verifying Copilot’s output and provides controls to delete history and disable personalization. Using these controls, plus feedback on problematic responses, is generally sufficient; there is no indication from the described behavior alone that someone else accessed the account or device.
For confidential or sensitive topics, avoid sharing data that should not be used as described in the Microsoft Privacy Statement, and rely on the privacy controls to manage what is stored and remembered.
References:
- Frequently asked questions about Copilot in the Clipchamp video player
- Privacy FAQ for Microsoft Copilot
- Revisit your Microsoft 365 Copilot Chat history
- Microsoft Copilot privacy controls
- Providing feedback about Microsoft Copilot with Microsoft 365 apps
- Frequently asked questions about Copilot in Forms
- Frequently Asked Questions about Copilot in Whiteboard