Securing the Internet of AI Agents
Amir Houmansadr
Abstract:
The Internet is undergoing a fundamental transformation as AI agents increasingly act on behalf of users, autonomously interacting with services, APIs, and online content. This emerging Internet of AI Agents introduces a new class of security and privacy challenges that span multiple layers, from the reasoning processes of large language models (LLMs), to the system-level behavior of agents, to their ecosystem-wide impact.
In this talk, I present a research agenda for securing the Internet in the age of AI agents. First, I show how the intelligence of AI agents (i.e., their LLM core) can itself become an attack surface, enabling a spectrum of security and privacy attacks that can be propagated at scale through the LLM ecosystem. Second, I demonstrate how AI agents introduce new system-level vulnerabilities, including the leakage of sensitive information through their network behavior, even under encrypted communication. Finally, I discuss how AI agents fundamentally reshape the security landscape, both by enabling new forms of large-scale attacks and by providing new opportunities for automated defense.
Together, these results highlight the need for a cross-layer approach to security and motivate a broader vision for building a secure and trustworthy Internet of AI agents.
Bio:
Amir Houmansadr is a Professor of Computer Science at the University of Massachusetts Amherst. He received his Ph.D. from the University of Illinois at Urbana-Champaign and was a postdoctoral researcher at the University of Texas at Austin. His research lies at the intersection of security, privacy, and AI-driven networked systems. He develops principled techniques for securing modern Internet systems, with a focus on trustworthy AI, privacy-enhancing technologies, and the analysis of real-world protocols and services. His work combines theory and practice, ranging from formal analysis using information theory and game theory to the design and deployment of systems with real-world impact. His research has been recognized with numerous awards, including the IEEE S&P Best Practical Paper Award, NSF CAREER Award, Google Faculty Research Award, DARPA Young Faculty Award, and DARPA Director’s Award.