Agentic AI systems - Promises, Risks and the Paths Forward

Nicole Obretincheva presented a review on Agentic AI systems, which includes an overview of work presented by Microsoft, OWASP, IBM, and the EU AI Act.

Abstract

Agentic AI systems are becoming increasingly widely used, and with that comes a new set of risks. In this talk, Nicole provides a brief overview of what these agentic AI systems are, the types of threats posed by agentic AI (drawing on work from OWASP and Microsoft), and how risk mitigation strategies are starting to take shape to address issues posed by agentic AI. The talk also explores how existing governance frameworks in the EU and the UK apply to these systems.