As AI evolves from static models to agentic systems capable of autonomous decision-making, the attack surface expands dramatically. This session explores how modern AI agents—powered by LLMs and external tools—can be exploited through prompt injection, tool manipulation, and data poisoning.

Through real-world demos and attack simulations, we will break down how attackers bypass safeguards, hijack AI workflows, and execute unintended actions across integrated systems. The talk also covers mitigation strategies including guardrails, context isolation, and secure agent design.

Attendees will leave with a practical understanding of AI attack vectors and defense mechanisms in modern production environments.