3.2.3. Responsible AI Principles
💡 First Principle: Microsoft's Responsible AI framework defines how AI should be built and used — not as a set of rules imposed on customers, but as a shared framework for building trustworthy AI systems. The six principles guide both how Microsoft develops Copilot and how organizations should govern AI use in their environments.
The principles are:
| Principle | What It Means | Practical Example |
|---|---|---|
| Fairness | AI should treat all people equitably, avoiding bias | Copilot responses shouldn't differ based on user demographics |
| Reliability and Safety | AI should perform consistently and safely under varied conditions | Copilot should not generate harmful content or fail unpredictably |
| Privacy and Security | AI should protect user data and respect privacy | Copilot doesn't use one user's data to train responses for others |
| Inclusiveness | AI should benefit all people and avoid exclusion | Copilot should be accessible to users with disabilities |
| Transparency | AI should be understandable — users should know they're interacting with AI | Copilot identifies itself as an AI and cites its sources |
| Accountability | People should remain responsible for AI decisions | Admins and users are accountable for how they use Copilot |
Applied to Copilot: confidently wrong output (hallucination) is a Reliability and Safety concern, mitigated by grounding and human review. Output that systematically disadvantages a group of people is a Fairness concern: evaluate outputs for bias and keep human judgment as the final decision. Citations let users check an answer against its sources (Transparency, supporting human oversight). Microsoft does not use prompts, responses or Microsoft Graph data from work accounts to train the foundation models (Privacy and Security). Microsoft's Transparency Notes document each AI system's intended uses, capabilities and limitations — supporting Transparency, and giving organizations what they need to deploy and govern it accountably. Whoever uses AI output stays accountable for it; that responsibility doesn't pass to the AI, its vendor or the admin who enabled it.
💡 Key Point: For the exam, focus on recognizing each principle by name and matching it to a scenario. "A company wants to ensure that Copilot doesn't make hiring decisions without human review" maps to Accountability. "Users should know when a response was generated by AI" maps to Transparency.
⚠️ Exam Trap: Responsible AI principles are not a compliance regulation or legal requirement imposed by Microsoft. They're a framework that Microsoft uses internally and shares as guidance. Organizations are responsible for governing their own AI use in accordance with applicable laws and their own policies.
Reflection Question: Your organization implements a policy that all Copilot-generated content in customer communications must be reviewed by a human before sending. Which Responsible AI principle does this policy most directly support?
