Agentic Networking: Securing AI Agents on Kubernetes - Haiyan Meng, Google & Evaline Ju, IBM

This talk focuses on "Agentic Networking" and its role in securing AI agents deployed on Kubernetes. Speakers Haiyan Meng from Google and Evaline Ju from IBM discuss how the unique behavior and communication patterns of AI agents challenge existing cloud-native infrastructure. They highlight the risks associated with prompt injections and poisoned tool responses, which can compromise agent decision-making and expose sensitive information. The session introduces adaptations in Kubernetes to manage traffic for AI-first protocols, exploring the extension of the Kubernetes Gateway API to ensure well-governed and auditable agent communication, while implementing security measures to protect agents in Kubernetes environments.