Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the cutting-edge challenges facing multi-agent security systems in this comprehensive lecture delivered by Christian Schroeder de Witt at the 2025 Cooperative AI Summer School. Delve into the complex landscape of AI security where multiple intelligent agents interact, examining the unique vulnerabilities and threats that emerge in distributed AI systems. Learn about high-dimensional anomaly detection techniques crucial for identifying malicious behavior in multi-agent environments, and understand how large language models can be leveraged for enhanced reasoning in security contexts. Discover the intricacies of multi-agent LLM post-training and its implications for developing emerging abilities that could revolutionize security protocols. Examine groundbreaking research on undetectable threats and perfectly-secure steganography, understanding how these concepts apply to modern AI systems where agents must cooperate while maintaining security. Gain insights into the intersection of AI policy and technical implementation, drawing from perspectives shared at high-level forums including the Munich Security Forum and specialized agent security convenings. Understand the collaborative research efforts between academic institutions and industry leaders like Google DeepMind, Microsoft, and OpenAI in addressing these critical security challenges. This lecture provides essential knowledge for researchers, practitioners, and policymakers working at the forefront of AI security, offering both theoretical foundations and practical considerations for securing multi-agent systems in an increasingly connected world.
