AI Security in Action: Hands-on cohort on Attacking & Defending AI apps & Agents

Direct & Indirect prompt injection attacks.

Jailbreak attacks on AI models and how to build defense guardrails.

Plan and execute AI red-team engagements using adversary simulation techniques.

Map AI/LLM applications against the OWASP LLM Top 10 to identify high-risk weaknesses.

Use MITRE ATLAS to design realistic attack scenarios and trace adversary tactics.

Craft and execute prompt injection, jailbreak, and agentic attacks to test system robustness.

Build simple data poisoning scenarios and explain their impact on model behavior and downstream systems.

Design and implement defensive patterns (input validation, output filtering, tool-use constraints) to mitigate these attacks.

Prototype LLM-based security scanners to detect injections, jailbreaks, manipulations, and risky behaviors.

Design and implement custom guardrails for input/output protection and integrate them into an existing AI application.

Run security benchmarking and focused penetration tests against LLM agents, producing clear, actionable findings.

Develop an incident response playbook for AI applications, services and agents.

Threat Modeling of AI applications and agents.

Build a practical AI SecOps workflow that covers the full AI/LLM supply chain (models, data, tools, plugins, infra).