Securing AI: Governance, Threat Protection, and Responsible AI on Azure

Topics Covered

• The evolving AI threat landscape
• Enterprise attack surfaces introduced by large language models (LLMs) and AI agents
• AI risk categories:
  • Data exposure
  • Prompt manipulation
  • Model abuse
  • Supply chain threats
  • Insider risks
• Mapping AI risks to cybersecurity and compliance frameworks
• Microsoft’s Secure Future Initiative and Responsible AI approach

Discussion

• Common security mistakes organizations make when deploying AI
• Lessons learned from early enterprise AI deployments

Topics Covered

• Applying Zero Trust principles to AI systems
• Identity-first security for AI workloads
• Entra ID and Entra Agent ID
• Securing model endpoints and inference services
• Private Endpoints and network isolation
• Data loss prevention strategies
• Key Vault and secrets management
• Securing Retrieval-Augmented Generation (RAG) architectures
• Governance of enterprise knowledge grounding

Architecture Review

Participants will walk through several reference architectures for:

• Internal enterprise copilots
• Customer-facing AI applications
• Autonomous AI agents
• Multi-agent systems

Topics Covered

• Governance frameworks for generative AI
• Model lifecycle governance
• Responsible AI controls
• Transparency and explainability
• Human-in-the-loop design
• Monitoring AI decision boundaries
• Agent oversight and approval workflows

Agent Governance Focus

• Governance and observability using Agent 365
• Agent inventory and lifecycle management
• Entra Agent ID for agent identity management
• Authorization boundaries for autonomous agents
• Risk management for agentic workflows
• Auditability and compliance reporting

Executive Discussion

How security leaders can enable innovation while maintaining governance, accountability, and regulatory compliance.

Topics Covered

• AI-assisted security operations
• Defender for Cloud security monitoring
• Sentinel investigation and threat hunting
• Detecting prompt injection attacks
• Monitoring agent behavior
• Visibility into model interactions
• AI-specific indicators of compromise
• Securing multi-agent environments

Demonstration

• Defender for Cloud security insights
• Sentinel AI-assisted investigations
• Detection of malicious prompt activity
• Agent behavior analytics

Hands-On Interactive Lab

Participants will assume the role of AI red team operators and evaluate a simulated enterprise AI assistant.

Red Teaming Scenarios

Exercise 1: Prompt Injection

Participants attempt to:

• Override system instructions
• Circumvent safety controls
• Manipulate model behavior

Exercise 2: Data Exfiltration

Participants explore methods to:

• Extract information beyond intended permissions
• Exploit retrieval and grounding weaknesses
• Test data segregation controls

Exercise 3: Agent Abuse

Participants evaluate:

• Unauthorized actions
• Excessive permissions
• Agent-to-agent trust failures
• Tool and plugin misuse

Exercise 4: Safety and Content Risk Testing

Participants test:

• Harmful content generation attempts
• Policy bypass techniques
• Content Safety effectiveness
• Prompt Shield protections

Lab Debrief

Participants document:

• Findings
• Risk severity
• Mitigations
• Governance recommendations
• Monitoring requirements

Closing Session: Building an Enterprise AI Security Program Topics Covered

• Governance operating model
• AI Center of Excellence considerations
• Policy recommendations
• Security architecture checklist
• Compliance readiness
• Metrics and KPIs for AI governance
• Executive reporting strategies

Deliverables

Participants receive:

• AI security and governance reference framework
• AI risk assessment checklist
• Agent governance maturity model
• AI red teaming playbook
• Recommended Azure security architecture patterns