AI Red Teaming: Offensive Security for AI Systems

As artificial intelligence becomes deeply integrated into modern software, securing these systems requires a fundamental shift from traditional web application testing. AI systems introduce unique attack surfaces, from hallucination exploits to compromised retrieval chains and unauthorized agent actions. This text-only course equips you with the foundational knowledge to think like an attacker and systematically pressure test AI implementations. You will transition from understanding basic security concepts to mapping out complex AI system boundaries and identifying critical vulnerabilities. What you'll learn: - Understand the fundamental architecture of AI systems and how their security risks differ from traditional software. - Identify and exploit prompt injection vulnerabilities and model hallucination risks. - Analyze retrieval-augmented generation (RAG) pipelines to detect data leakage and unauthorized data blending. - Evaluate agentic tool integrations to prevent unauthorized action execution and privilege escalation. - Map the AI system boundary and model potential threats across the entire data supply chain. - Apply modern security frameworks, including the OWASP Top 10 for LLMs, to structured vulnerability assessments. The curriculum flows logically from foundational AI security definitions and terminology to practical threat modeling scenarios. Through structured text-based explanations and written analysis exercises, you will explore real-world attack vectors and defense strategies. This course is designed for cybersecurity beginners, traditional penetration testers, and developers looking to enter the AI security space, with no prior AI red teaming experience required. Begin your journey into offensive AI security and learn how to defend the next generation of applications.