Red Teaming AI: Proactive Security in the Age of Artificial Intelligence

No ratings

Presented at SecTor 2024 by

Jeremy Richards

As artificial intelligence systems become increasingly prevalent and powerful, ensuring their security and robustness is paramount. This talk explores the critical practice of red teaming in AI development and deployment, drawing insights from MITRE ATLAS™, NVIDIA's AI Red Team approach, and recent vulnerability reports from initiatives like Protect AI's huntr program. We'll dive into the methodologies, challenges, and ethical considerations of simulating adversarial attacks on AI systems. From uncovering potential vulnerabilities to strengthening defenses across the AI lifecycle, attendees will gain insights into how red teaming contributes to building more secure and trustworthy AI. Real-world case studies, including recent critical vulnerabilities in popular AI/ML tools, practical strategies for implementing red teaming in various AI contexts, and emerging topics in AI security will be discussed, equipping participants with valuable knowledge to enhance their own AI security practices.