AWS Nitro Sandboxes brings hardware-enforced isolation to pods on Amazon EKS, aiming to avoid the usual choice between sharing efficient Kubernetes nodes and assigning sensitive workloads dedicated EC2 instances. The concern is that ordinary containers share a host kernel, so a container escape can threaten neighboring workloads. The risk is especially pressing for AI agents that execute generated code, SaaS platforms serving many customers on shared clusters, privileged CI/CD jobs, and regulated organizations that need evidence of protection. Each sandboxed pod instead runs in a lightweight virtual machine with its own operating system, separated from the host and other pods by the AWS Nitro System.
The feature is designed to fit existing EKS workflows: on EKS Auto Mode, a pod-spec change enables sandboxing, while container images and common tools and integrations - including Pod Identity, storage, CloudWatch, networking, and kubectl commands - continue to work. AWS says the system passes Kubernetes conformance tests; standard EKS requires a custom AMI. Nitro Sandboxes also aims to make shared-node performance predictable: CPU and memory are isolated, while configurable throttles limit shared resources such as network bandwidth, disk I/O, DNS, logging, and connection tracking. Two modes are described. Isolated Mode protects the host and neighboring pods; Confidential Mode additionally treats the machine operator as untrusted, using cryptographic attestation and AWS KMS policies to control decryption and provide evidence of the system image and policy running in the sandbox.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.