Cloud Security Cloud Workload Security

Rafay and Fortanix Bring Confidential AI Workloads to Kubernetes Pods

Rafay Systems and Fortanix have expanded their confidential AI offering from virtual machines to Kubernetes pods, combining the Rafay Platform with Fortanix Confidential AI and the CNCF Confidential Containers project. Announced on Nov. 12, 2026, the integration is intended to let neoclouds, sovereign AI cloud providers and enterprises offer confidential AI workloads as a service.

The change brings the companies’ existing confidential-computing integration into a Kubernetes deployment model. Rafay’s platform orchestrates workloads running on Confidential Containers, also known as CoCo. The project, which is incubating at the Cloud Native Computing Foundation, is built on Kata Containers and runs each pod inside its own lightweight confidential virtual machine.

In this design, the CPU encrypts the pod’s memory. On supported hardware, GPUs can also be included in the confidential computing trust boundary. That hardware-backed isolation is intended to protect workload data while it is being processed, a gap that conventional encryption does not address by itself.

Attestation governs access to encryption keys

Fortanix Confidential Computing Manager performs composite attestation across CPU and GPU trusted execution environments, as well as firmware and software components. The process checks whether the environment running a workload meets the conditions required for access to protected data.

Fortanix Data Security Manager releases encryption keys only to environments that pass attestation. If a workload fails that check, its keys remain protected and it cannot access the sensitive data those keys safeguard. The arrangement makes attestation a gate for data access rather than relying only on the presence of encrypted memory.

Fortanix Chief Product Officer Anuj Jaiswal said model owners, enterprises and operators want assurance that other parties cannot see what they own. He said Fortanix Confidential AI releases keys only to attested workloads. Rafay Chief Product Officer Mohan Atreya said the move to Confidential Containers takes the model to the Kubernetes pod, where AI teams already deploy workloads.

Rafay adds cluster and tenant operations

Rafay provides the operating layer for deploying and managing the setup. Its blueprints install Fortanix CoCo add-ons and keep them synchronized across confidential clusters. Operators onboard GPU infrastructure once and can then provide isolated tenants to customers, according to the announcement.

Administrators can package a confidential workload as a SKU in Rafay SKU Studio. End users can launch that workload on demand through the Rafay Developer Hub, rather than submitting a support ticket. Policy, access controls and quotas apply across tenants and clusters.

These functions place provisioning and governance alongside the underlying confidential-computing controls. The announcement describes a service model in which an operator prepares cluster capabilities and workload offerings, while customers select and launch an available workload. It does not specify supported hardware models, Kubernetes versions, performance figures or pricing.

Support is available now

Rafay support for CoCo-based deployments with Fortanix Confidential AI is available now, alongside the companies’ existing confidential VM integration. Fortanix had announced Confidential AI support for CoCo on Oct. 19. The latest announcement extends that work through Rafay’s Kubernetes workload orchestration and service-provisioning features.

The companies also cited research by Fortanix and Atomik Research on AI data protection. The study found that 93% of AI leaders understand traditional encryption does not protect data while AI processes it, while only half said they can fully verify that protection is in place during inference. The findings underscore the distinction the offering targets: keeping data encrypted at rest or in transit is different from protecting it while a model is processing it.

For organizations deploying AI through Kubernetes, the integration combines pod-level confidential execution, attestation-controlled key release and a platform for onboarding tenants and launching workloads. Its stated availability gives cloud operators and enterprises an option to extend confidential AI deployments beyond virtual machines into Kubernetes pods.