Purple Teaming Agentic Systems
Built around the newly updated OWASP Top 10 for LLM Applications, it focuses on the security problems that actually show up when organizations ship AI and agentic systems, not frontier lab headlines.
🔁 Every lab follows the same purple team loop:
🔴 Break: exploit a flawed AI implementation
🔵 Fix: apply mitigation and remediation at the root cause
✅ Validate: confirm the fix holds up against the original attack
🧭 What you will cover:
Theory and Intro (30 min): the updated OWASP LLM Top 10 and how AI risk differs from, and overlaps with, classic application security
Lab I: Direct and Indirect Prompt Injection (45 min): exploit both injection types and test how well current mitigations hold up
Lab II: Downstream LLM Output Attacks (45 min): attack LLM systems connected to downstream components, then remediate the weak points
Lab III: Excessive Agency and the Confused Deputy Problem (45 min): abuse over-permissioned agents and lock them down
Lab IV: MCP and Tool Attacks (45 min): vulnerable and malicious MCP configurations, malicious skills, and weaknesses across the wider agentic ecosystem, plus remediation
Lab V: Data-Oriented AI Attacks (30 min): RAG pipeline attacks and data poisoning, with matching remediations
🎯 What you will walk away with:
A practical understanding of the real risks in AI and agentic implementations
Hands-on experience both attacking and securing AI systems
A repeatable break, fix, validate method you can apply at work
Certificate of participation with CPE credits
👥 Who this is for:
Penetration testers expanding into AI security
Engineers and security professionals responsible for implementing AI or agentic systems in their organization
📋 Prerequisites:
Solid knowledge of application security fundamentals and security principles
Hands-on application security experience
Familiarity with Burp Suite and common classes of web application vulnerabilities
👨💻 Instructor:
Jacob Simmons, Co-Founder, Aphotic Security
Jacob Simmons is the Co-Founder of Aphotic Security and an OSCP-certified cybersecurity professional and ethical hacker with a BEng in Computer Engineering. He moved into offensive security to help organizations counter evolving digital threats, combining penetration testing and cloud security expertise with hands-on, real-world experience. Jacob has presented research at BSides Cape Town, BSides Johannesburg, and the AISA conference, and recently delivered a webinar on practical AI security, focused on separating real agentic security problems from industry hype. When he is not breaking AI systems, he enjoys finding new ways to hack cloud environments.