3.5. Reflection Checkpoint

Key Takeaways

Before proceeding, ensure you can:

• Identify and distinguish fabrication, prompt injection, and over-reliance in scenario descriptions
• Select appropriate verification steps (citation check, human review, cross-referencing) for different content types
• Explain why Copilot can surface sensitive data even without breaking permission boundaries
• Describe how sensitivity labels and DLP policies restrict Copilot outputs (not just inputs)

Connecting Forward

Phase 4 shifts from risks and safeguards to active skill-building: crafting effective prompts, managing conversations, and using the Copilot prompt management features the exam heavily tests. The responsible AI principles from this phase remain relevant — effective prompting includes knowing when to add verification steps to your workflow.

Self-Check Questions

1. A colleague argues that prompt injection only affects IT professionals building AI systems, not regular business users. How would you correct this, using a concrete business scenario?
2. Your organization's legal team has labeled all contract documents as "Highly Confidential." When you ask Copilot to draft a contract summary for an external partner, you receive a restricted content warning. Is this a Copilot error? What should you do?