AI security and privacy, agentic systems, backdoors, multimodal attacks, differential privacy.
Eugene Bagdasarian co-leads the UMass AI Security Lab. His group studies security and privacy attacks on AI systems used in the real world, including autonomous agents, generative models, and federated learning. Work includes Contextual Integrity for agent privacy (AirGapAgent), policy checks on agent tool use, prompt leakage, backdoors in federated learning and diffusion models, and adversarial attacks on multimodal embeddings. He also works on differential privacy, fairness, and use-based privacy. He is a part-time senior research scientist at Google on agentic privacy and security, and co-leads the UMass AI Safety Initiative.
Eugene Bagdasarian is an Assistant Professor of Computer Science at UMass Amherst CICS. He completed his Ph.D. at Cornell Tech. Before graduate school he earned an engineering degree from Bauman University and worked as a software engineer at Cisco. His research has received an Apple Scholars in AI/ML Fellowship and a USENIX Security Distinguished Paper Award.
He co-leads the UMass AI Security Lab.