The My Account portal allows users to delete their Microsoft 365 Copilot activity history (which includes both input prompts and AI responses), but it does not provide a feature to review or read past interactions. Viewing interaction data is restricted to the respective Copilot apps or to administrators using Microsoft Purview eDiscovery. Furthermore, deleting the activity history only purges the chat interaction logs; it does not delete or modify any associated files, Word documents, OneNote pages, or notebooks where the user may have actively saved the Copilot-generated content.

Microsoft Learn: "Data, Privacy, and Security for Microsoft 365 Copilot" – Section: Deleting the history of user interactions with Microsoft 365 Copilot. This confirms users can delete prompts/responses via the My Account portal, while viewing requires admin tools like Purview.

Microsoft Support/Learn Documentation: "Delete your Microsoft 365 Copilot activity history" – Details the scope of deletion, confirming that standalone files and saved content outside the interaction log remain unaffected.

Enterprise Data Protection (EDP) is the specific privacy and security commitment in Microsoft 365 Copilot that guarantees organizational data remains secure. When users access Copilot using their work or school accounts (Microsoft Entra ID), EDP ensures that user prompts, generated responses, and any business data accessed via Microsoft Graph remain strictly within the Microsoft 365 tenant's trust boundary. Crucially, EDP guarantees that this business data is never logged or used to train Microsoft's underlying foundation large language models (LLMs). While Zero Trust and Sensitivity Labels are important security features within the ecosystem, EDP is the exact framework explicitly responsible for preventing customer data from being used in AI model training.

Microsoft Learn: "Data, Privacy, and Security for Microsoft 365 Copilot" - Section: Enterprise data protection (EDP). Details how prompts, responses, and data accessed through Microsoft Graph are not used to train foundation LLMs. (https://learn.microsoft.com/en-us/microsoft-365-copilot/microsoft-365-copilot-privacy)

Microsoft Learn: "Enterprise data protection in Copilot for Microsoft 365 and Microsoft Copilot" - Section: How does EDP protect your data?. Confirms that customer data is protected and not used to train the AI foundation models. (https://learn.microsoft.com/en-us/copilot/enterprise-data-protection)

When users navigate to the Microsoft 365 My Account portal (under Settings & Privacy) and select the "Delete history" option for their Copilot activity, the system performs a bulk deletion. It removes the user's entire Microsoft 365 Copilot interaction history. Microsoft's privacy controls for this specific portal action do not currently allow for granular deletions, such as filtering by a specific date range, selecting a single conversation, or isolating interactions with a specific agent. Selecting this option clears all saved prompts and responses associated with that user's account.

Microsoft Support Documentation: "Delete your Microsoft 365 Copilot activity history." Section: How to delete your Copilot activity history. (Details the exact steps in the My Account portal and confirms it deletes the entire history).

Microsoft Learn: "Data, Privacy, and Security for Microsoft 365 Copilot" (Article ID: microsoft-365-copilot-privacy). Section: Copilot activity history. (Outlines user data retention and the privacy portal's bulk deletion capabilities).

In Microsoft 365 Copilot, the instructions field serves as the core system prompt that dictates the agent's behavior, tone, and persona. To configure an agent to communicate in a "warm and friendly" manner, you must explicitly define this stylistic preference within the instructions. The other options serve different functions: the "description" informs users about the agent's purpose to aid in discovery, "capabilities" determine the agent's functional reach (such as web search or utilizing specific plugins), and "suggested prompts" are merely clickable conversation starters provided to the end user.

Microsoft Learn: "Overview of declarative agents for Microsoft 365 Copilot"

Section: "Core components of a declarative agent" (Instructions subsection)

Details: Official documentation specifies that the Instructions property provides specific directions to the foundation model to shape its persona, behavior, and response tone.

Link: https://learn.microsoft.com/en-us/microsoft-365-copilot/extensibility/overview-declarative-agent

Microsoft Copilot Studio Documentation: "Quickstart: Create and deploy a Copilot"

Section: "Instruct your copilot"

Details: Details how developers use the Instructions text box to give the copilot a specific personality, dictate its conversational tone, and establish operational guardrails.

Link: https://learn.microsoft.com/en-us/microsoft-copilot-studio/nlu-gpt-quickstart

When you use Microsoft 365 Copilot, particularly under the umbrella of commercial data protection, user prompts and AI-generated responses are not retained by Microsoft. If you start a temporary chat and then initiate a new chat (or close the browser), the previous session's context, prompts, and responses are deleted immediately from the system's short-term memory. They are not saved to OneDrive, moved to a notebook, or cached for an hour. This behavior enforces Microsoft's strict privacy commitments ensuring that commercial data is not saved, Microsoft has no "eyes-on" access to your chat history, and your session data is never used to train the underlying foundational Large Language Models (LLMs).

Microsoft Learn - Official Documentation: Data, Privacy, and Security for Microsoft Copilot for Microsoft 365. Section: "How Copilot protects organizational data." This document specifies that prompts and responses are not retained or stored after a session is cleared or closed.

Microsoft Learn - Official Documentation: Microsoft Copilot Privacy and Protections. Section: "Commercial Data Protection." Explicitly outlines that when a user ends a chat session, the chat data is discarded immediately to ensure enterprise security and privacy standards.

In Microsoft 365 Copilot, you can use a page (specifically through the Copilot Pages feature) to save, edit, and collaborate on AI-generated content. When you generate a report or gather information using Copilot or its specific agents in Business Chat (BizChat), transitioning the response to a Page transforms it from an ephemeral chat message into a durable, editable canvas. This allows you to refine the text, format the report, and optionally share it with other users for real-time, multiplayer collaboration within the Microsoft 365 ecosystem.

Microsoft 365 Blog & Official Announcements: "Microsoft 365 Copilot Wave 2: Pages, Python in Excel, and agents" (September 2024). Details the introduction of Copilot Pages as a persistent canvas designed specifically for saving, editing, and collaborating on Copilot-generated responses.

Microsoft Learn Documentation: "Overview of Microsoft 365 Copilot Business Chat (BizChat) and Pages" - Highlights the workflow of taking conversational outputs and moving them to a Page for durable editing and sharing.

When a declarative agent, such as a custom "Researcher agent," is published and accessed from within the Microsoft 365 user interface, it is in a deployed state for end-user consumption. The Microsoft 365 client does not provide authoring controls to alter the agent's foundational behavior. You cannot change its base instructions, add new starter prompts, or delete existing suggested prompts directly from the M365 chat interface. To make any structural or behavioral modifications, the developer or administrator must return to the original authoring environment (e.g., Microsoft Copilot Studio), edit the configuration parameters there, and republish the agent.

Microsoft Learn Documentation: Overview of declarative agents for Microsoft 365 Copilot. Section: "Authoring and Publishing". This document specifies that the creation and modification of declarative agent configurations (instructions, prompts, knowledge) occur exclusively within Copilot Studio or via the Teams Toolkit, not the M365 client UI.

Microsoft Learn Documentation: Manage Microsoft Copilot Studio agents in Microsoft 365. Section: "Edit an existing agent". It dictates that to update an agent deployed to M365, makers must open the agent in Copilot Studio, make the necessary changes, and select "Publish" again.

Microsoft 365 Agent Builder natively supports using Microsoft Teams data—including channels, group chats, and 1:1 chats—as a direct knowledge source to ground an agent. However, past Microsoft 365 Copilot interaction transcripts are not an officially supported data source; grounding is restricted to enterprise sources like SharePoint, Dataverse, web URLs, Graph Connectors, and Teams. Furthermore, image files such as JPG or PNG cannot be used. Copilot specifically parses text content for knowledge retrieval, explicitly restricting supported file uploads to standard document formats such as .docx, .pdf, .pptx, .xlsx, and .txt.

Microsoft Learn, Add knowledge sources to your declarative agent in Microsoft 365 Copilot (Article: microsoft-365-copilot/extensibility/agent-builder-add-knowledge), Section: "Microsoft Teams data".

Microsoft Learn, Agent Registry in the Microsoft 365 admin center (Article: microsoft-365/admin/manage/agent-registry), Section: "Supported file types and limits" (Confirming the exclusion of image formats).

Microsoft Learn, Knowledge sources summary - Microsoft Copilot Studio (Article: microsoft-copilot-studio/knowledge-copilot-studio), Section: "Supported knowledge sources" (Confirming the exclusion of Copilot conversations).

Overreliance occurs when an individual trusts an AI system's output without applying critical thinking, domain expertise, or independent verification. In this scenario, the merchandiser accepted Copilot's suggestions as fact and acted on them without reviewing them against other data sources like market trends, which is a direct example of this responsible AI risk. Human oversight is a critical safeguard against the potential inaccuracies or biases in AI-generated content.

A. verification: This is the process of checking for accuracy, which is what the merchandiser failed to do. It is the opposite of their action.

C. fabrication: This refers to the AI generating false or nonsensical information (hallucinating). While the AI's output might be flawed, the user's action is the focus here.

D. prompt injection: This is a security attack where a user crafts a prompt to make the AI bypass its safety features or intended functions, which is not what happened.

1. Microsoft. (2023). Microsoft Responsible AI Standard

v2. Section 4.2.1

"Human-AI interaction patterns

" discusses the risk of automation bias and overreliance

stating "Overreliance on the AI system can lead to an inability to detect or correct system errors."

2. Microsoft Learn. (2024). Identify responsible AI principles. In "Describe the capabilities of Microsoft Copilot for Microsoft 365

" Module 1. This module emphasizes the principle of "Accountability

" where users are accountable for the final output and should review AI-generated content.

Users can manage their Microsoft 365 Copilot interaction history, including deleting specific conversations, through the My Account portal. This portal provides user-level control over personal data without needing administrator intervention. Deleting history here removes it from the user's view and from being used to refine Copilot's responses. The other options are for administrative tasks or do not offer the required data management functionality for an individual user.

A. The Microsoft 365 Copilot app is for interacting with Copilot, not for managing the underlying conversation data history.

B. The Microsoft 365 admin center is for tenant-wide administrative tasks and requires elevated permissions, which contradicts the user-level requirement.

C. The Microsoft Purview compliance portal is for administrators to manage eDiscovery, retention, and compliance, not for individual user data deletion.

1. Microsoft Learn. "Manage Microsoft Copilot for Microsoft 365 interaction history." Microsoft 365 Documentation. Section: "Delete your Copilot for Microsoft 365 interaction history." This document states

"To delete your Copilot for Microsoft 365 interaction history

you can go to the My Account portal."

2. Microsoft. "Data

Privacy

and Security for Microsoft Copilot for Microsoft 365." Microsoft 365 Documentation. Section: "User Controls." This document outlines that users have controls over their data

including the ability to delete their prompt history via their account settings.