Your Models Are Vulnerable: How KitOps Turns KServe Into a Zero-Tru... Brad Micklea & Gavrish Prabhu

This talk explores how KitOps enhances KServe by transforming it into a zero-trust inference platform. Speakers Brad Micklea and Jozu & Gavrish Prabhu delve into what it takes to ensure model integrity against threats like poisoning and tampering, using ModelKits to package models securely with cryptographic signatures stored in existing registries. The session reveals critical ML-pipeline attack vectors and practical production patterns to mitigate them, including the use of init containers for model signature verification and OPA policies to block unattested models. Attendees will gain access to proven manifests, real-world demos, and a checklist for secure deployment, enabling them to strengthen KServe with familiar tools.