AI-agents veilig inzetten: hoe je autonome systemen beschermt tegen aanvallen

Sriram Madapusi Vasudevan examines security challenges when deploying autonomous AI agents in production environments. The presentation identifies vulnerabilities within the ReAct loop—spanning context handling, reasoning processes, and tool execution—and outlines mitigation strategies including defense-in-depth approaches and LLM-based validation systems to prevent threats such as memory