A hands-on walkthrough of building a zero trust AI pipeline for healthcare using Red Hat OpenShift confidential containers (CoCo) and RHEL confidential virtual machines (CVMs). The scenario centers on deploying a PHI de-identification NLP model (obi/deid_roberta_i2b2) while enforcing strict security policies via hardware attestation. The architecture separates a trusted on-premises cluster (running Red Hat build of Trustee, OpenShift Pipelines, and GitOps/Argo CD) from an untrusted cloud deployment cluster on Azure Red Hat OpenShift. Three automated pipelines handle the full lifecycle: a standard build pipeline, a CoCo-fy pipeline that hardens containers with sealed secrets and mTLS sidecars, and a CVM pipeline that deploys AMD SEV-SNP-backed instances on Azure and AWS. Secrets are only released after hardware attestation succeeds, and developers like Janine need no changes to their existing workflow.
