Fabrication
A fabrication is a plausible-sounding but incorrect or invented output from a generative model. AB-730 covers fabrications as a risk and the verification steps that mitigate them.
A fabrication is a plausible-sounding but incorrect or invented output from a generative model. AB-730 covers fabrications as a risk and the verification steps that mitigate them.
Microsoft 365 Copilot is the generative AI assistant built into Microsoft 365 apps such as Word, Excel, Outlook, and Teams.
A prompt is the natural-language instruction a user gives Copilot to produce a result.
An agent is a customizable Copilot assistant configured with knowledge and instructions to handle specific tasks.
The Agent Store is the catalog where users add prebuilt agents instead of building a new one.
Researcher is a Microsoft 365 Copilot agent that performs multi-step research across work and web sources to produce in-depth answers.
Analyst is a Microsoft 365 Copilot agent that reasons over data to produce analysis and insights for business decisions.
Grounding is how Copilot bases its responses on relevant context such as your work files, web data, or the current app.
Microsoft Graph is the data layer that exposes your Microsoft 365 content to Copilot so responses reflect your organization's information.
Copilot Pages is a collaborative canvas where teams work with Copilot-generated content together in real time.
A notebook in Microsoft 365 Copilot is a workspace that groups prompts, sources, and conversations around a topic.
Prompt injection is an attack that plants malicious instructions in content a model reads so it behaves against the user's intent.
A fabrication is a plausible-sounding but incorrect or invented output from a generative model.
Responsible AI is the set of principles and practices for using AI safely, fairly, and transparently.
A sensitivity label is a classification applied to content that enforces protection and restricts how Copilot can use it.