Agentic AI Runtime Security Workshop

A hands-on workshop focused on securing AI agents at runtime through governance, policy enforcement, access control, threat containment, human oversight, isolated execution, telemetry, and security control verification.

Part I — Theory Session

Session 1 — Why Agentic AI Needs Runtime Governance

Focus: Why agentic AI creates a new class of security and governance problems.

• AI has moved from answering to acting
• Why agentic AI creates a new security problem
• What ISO/IEC 42001 brings
• From governance expectations to runtime enforcement
• Selected ISO/IEC 42001 controls for the workshop
• Core problems Runtime Shield is trying to solve

Session 2 — Runtime Shield Architecture

Focus: How Runtime Shield places a control layer between the AI agent and real systems.

• What Runtime Shield is
• End-to-end Runtime Shield architecture
• The Runtime Shield Gateway as the main interception point
• What the gateway evaluates
• Possible runtime outcomes

Session 3 — Case Studies: Identity, Access and Manipulation

Focus: How Runtime Shield responds to common runtime security risks.

• Baseline Story — Anika and Mira: one normal working day with access to files, tools and APIs
• Case 1 — Restricted Resource Access
• Case 2 — Prompt Injection
• Case 3 — Authorization Drift

Session 4 — Human Oversight, Safe Execution and Evidence

Focus: What happens when agent actions become higher risk and how organizations prove what happened.

• Case 4 — High-Risk Autonomous Action
• Case 5 — Incident Investigation

Session 5 — Verification, Assurance and Continual Improvement

Focus: Why controls must be tested and continuously improved, not simply defined.

• Why verification matters
• Control design vs control effectiveness
• Continual improvement of runtime controls

Part II — Hands-On Lab Session

Lab 1 — Role-Based Access Control

Test how Runtime Shield enforces role-based permissions when an AI agent attempts to access restricted resources.

Lab 2 — Prompt Injection & Threat Containment

Observe how unsafe tool requests triggered by manipulated content are intercepted and contained at runtime.

Lab 3 — Authorization Drift & Re-Evaluation

See how changed roles or permissions trigger a fresh authorization decision instead of reusing stale access.

Lab 4 — Human-in-the-Loop & Isolated Execution

Trigger a high-risk action, route it for human approval, and observe controlled execution in an isolated environment.

Lab 5 — Telemetry, Auditing & Incident Investigation

Use runtime telemetry to reconstruct who did what, which policy decision was made, and whether an action executed.

Lab 6 — Security Control Verification

Run repeatable security checks to compare expected and actual outcomes and verify that controls remain effective.

Participant Takeaways

Participants Will Gain

• Understanding of Agentic AI runtime security risks
• Practical knowledge of runtime governance and policy enforcement
• Understanding of role-based access control and least privilege
• Experience with prompt injection containment
• Understanding of authorization drift and re-evaluation
• Experience with Human-in-the-Loop security controls
• Exposure to isolated and sandboxed agent execution
• Experience with telemetry, auditing and investigation
• Understanding of security control verification
• Practical experience through hands-on attack and defense scenarios

Workshop Details

Date: Saturday, 31 October 2026

Time: 10:30 AM – 01:30 PM IST

Duration: 3 Hours

Format: Theory + Hands-on Labs

Certificate: Certificate provided to participants

Certificate

Certificate Provided to Participants

Participants who attend the workshop will receive a certificate recognizing their participation in the Agentic AI Runtime Security Workshop.