Who Else Can See Your Prompt? Confidential AI from Chip to Agent

AI no longer just helps debug our code. It now handles our most sensitive data: patient records, financial models, citizen case files and the context AI agents carry between systems. Encrypting data at rest and in transit is no longer enough. The real gap is what happens while the data is being processed. This session shows how confidential computing closes that gap. It covers hardware-based trusted execution environments on CPUs and GPUs, and attestation, which proves an environment is trustworthy before any keys, models or data are released. It also walks through deployment options, from confidential VMs and containers to rack-scale GPU systems like NVIDIA Vera Rubin NVL72. Using scenarios from healthcare, finance and the public sector, you'll learn how to protect both your data and your model IP, and where confidential AI fits alongside sovereignty and compliance.