Skip to content

Implement AI Solutions Responsibly

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers implementing responsible AI practices in Azure AI Foundry solutions. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts

Implement Content Moderation Solutions

📖 Docs: Azure AI Content Safety overview

Overview

  • Content moderation ensures AI solutions do not produce harmful or unsafe outputs
  • Azure AI Content Safety detects and classifies:
    • Hate
    • Violence
    • Sexual content
    • Self-harm

Key Points

  • Supports both text and image moderation
  • Categories are scored on severity levels
  • Can block, review, or log flagged content

Exam Tip

If the scenario involves filtering harmful content, use Azure AI Content Safety

Configure Responsible AI Insights, Including Content Safety

📖 Docs: Responsible AI dashboard

Overview

  • Responsible AI Insights provide tools to evaluate fairness, explainability, and safety
  • Content Safety integration provides monitoring and reporting
  • Helps track risks and compliance with Responsible AI standards

Key Points

  • Dashboards show model performance across sensitive attributes
  • Identify and mitigate bias during testing
  • Can integrate with CI/CD for continuous monitoring

Best Practices

Use Responsible AI dashboards during both training and deployment

Implement Responsible AI, Including Content Filters and Blocklists

📖 Docs: Content filtering in Azure OpenAI

Overview

  • Azure OpenAI includes content filters for harmful outputs
  • Custom blocklists can be applied to prevent specific terms or topics

Key Points

  • Filters cover categories like sexual, violent, self-harm, hate speech
  • Filters are configurable to allow stricter enforcement
  • Blocklists help align outputs with organizational policies

Exam Tip

Remember the difference: - Content filters = built-in moderation - Blocklists = custom word/phrase restrictions

Prevent Harmful Behavior, Including Prompt Shields and Harm Detection

📖 Docs: Prompt engineering and safety

Overview

  • Harmful behaviors include prompt injection and jailbreaking
  • Mitigation strategies:
    • Prompt shields: predefined templates that reduce manipulation risks
    • Harm detection: automatic monitoring of inputs and outputs

Key Points

  • Validate and sanitize user inputs
  • Avoid exposing raw system prompts
  • Monitor logs for malicious usage attempts

Exam Tip

Watch for scenarios describing prompt injection attacks — answer with prompt shields or harm detection

Design a Responsible AI Governance Framework

📖 Docs: Cloud Adoption Framework: Govern AI

Overview

  • Governance ensures Responsible AI is part of the lifecycle of AI projects
  • Includes:
    • Policies
    • Processes
    • Tools for oversight

Key Points

  • Aligns with Microsoft’s 6 principles: fairness, reliability, privacy, inclusiveness, transparency, accountability
  • Requires collaboration between technical and compliance teams
  • Governance includes monitoring, incident response, and audits

Use Case

Enterprise deploying AI copilots with oversight committees and auditing tools

Quick‑fire revision sheet

  • 📌 Use Azure AI Content Safety for text/image moderation
  • 📌 Responsible AI Insights = dashboards for fairness and safety
  • 📌 Content filters = built-in moderation, blocklists = custom restrictions
  • 📌 Prompt shields + harm detection defend against malicious inputs
  • 📌 Governance framework aligns with Microsoft’s 6 Responsible AI principles