Title:

Deus Ex Machina: Emerging Opportunities and Perils in an Agentic World

Keynote speaker:

Dr. Ioannis Agadakos, Montclair State University, USA

Dr. Ioannis Agadakos is an Assistant Professor of Cybersecurity in the School of Computing at Montclair State University. He earned his MSc and PhD in Computer Science from Stevens Institute of Technology (2017, 2019), and a B.Eng. in Electronic and Computer Engineering and an MSc in Embedded Systems from the Technical University of Crete, Greece (2010, 2013). Before joining academia, Dr. Agadakos held prominent industry and research positions, including Software Engineer on the AWS QuickSight Machine Learning Insights team, Applied Scientist on Amazon Web Services’ Privacy and Security Automation team, and Computer Scientist in the Diverge Lab led by Professor Robertson at Northeastern University’s Khoury College of Computer Sciences, where he worked on the emerging field of neural binary analysis. Prior to that, he spent three years at SRI International as an Advanced Computer Scientist in the Critical Infrastructure Group, leading research in IoT security and device fingerprinting and serving as a subject-matter expert for DHS and other government agencies. Dr. Agadakos has authored over 20 peer-reviewed publications in premier venues including RAID, ACSAC, MobiSys, IEEE EuroS&P, and Transactions on Security and Privacy (TOPS). He holds two US patents, with two additional applications pending. His current research focuses on the security of agentic AI systems, IoT platforms, privacy-preserving frameworks, and software hardening. Dr. Agadakos’s work spans both defensive techniques and the discovery and mitigation of privacy and security threats in complex systems.

Abstract:

AI systems have reshaped numerous scientific and industrial domains in recent years, advancing the state of the art in image recognition, speech transcription, medical decision support, and recommendation systems. With the advent of transformer architectures and large language models (LLMs), however, we have crossed a new threshold: these systems are no longer just specialized tools, but general-purpose engines that are already transforming how many of us work, communicate, and create—and will continue to reshape most professions for years to come.

LLMs now form the foundation of agentic AI: systems that can perceive, reason, and act on our behalf. These agents are rapidly gaining traction not only in professional environments, where organizations increasingly expect employees to use AI to remain competitive, but also among casual users who rely on them for everyday tasks. Their appeal rests on three main pillars: (1) a conversational interface that lets people interact in natural, everyday language; (2) rich multimodal input, where a user can, for instance, upload a photo of a diseased plant and simply ask what might be causing the leaf discoloration—questions that would be far more difficult to formulate as traditional search queries; and (3) the ability to connect and reason over heterogeneous data sources and modalities, enabling forms of analysis that were previously difficult or impractical with conventional systems.

In this keynote, I will explore how agentic AI can augment human expertise, using as a case study an interactive, agent-assisted workflow for tackling a traditionally hard and tedious problem: identifying code similarity. I will discuss how agentic technologies are giving rise to novel programming patterns in which LLMs are treated as core programming primitives and embedded directly into software in place of, or alongside, conventional logic. At the same time, I will examine how the very strengths of these systems can become liabilities—amplifying risks to user privacy, enabling new forms of misinformation, boosting the capabilities of malicious actors, and opening unexpected pathways for longstanding software attacks. In this emerging landscape, agents may not only help us build the future, but also quietly embed backdoors in code, distort facts at scale, and reshape our information ecosystem in ways we do not yet fully understand.

Keywords: Large language models, Agentic AI systems, Software and application security, Program analysis and code similarity, Privacy and misinformation, Adversarial machine learning, Automated Reasoning