AWS Nitro Sandboxes give each EKS pod or ECS task its own Nitro-isolated virtual machine
Amazon Web Services announced AWS Nitro Sandboxes on 9 November, lightweight virtual machines managed by the Nitro Hypervisor that can wrap an Amazon EKS pod, an Amazon ECS task or a standalone VM. Each workload gets the isolation boundary of its own Nitro virtual machine while sharing a single EC2 instance with others, which AWS pitches for untrusted, multi-tenant or regulated code, including AI agents and code generated by language models.
According to the announcement, the sandboxes run on customers' own EC2 instances with performance close to native EC2 and inherit the hypervisor's isolation. Workloads can be checked by cryptographic attestation, so that secrets are released only to a sandbox that has proved its integrity. In EKS and ECS, AWS describes enabling the feature as a one-line change; each pod or task then gets its own VM, and the usual integrations, Pod Identity, the VPC CNI, EBS, EFS and S3 storage and CloudWatch, continue to work.
Customers who want full control can run Nitro Sandboxes directly on EC2 with their own machine image, orchestration and virtual machine monitor. Amazon Linux and Firecracker are supported natively, and AWS says patches for other options are available. The feature is offered on most Nitro-based instance types in all commercial regions at no additional charge; details are in the EKS, ECS and EC2 user guides.

Why it matters
Containers on one host share a kernel, so running code you did not write in a pod has meant either trusting that boundary or building a VM layer yourself. Getting a per-pod VM inside an existing cluster with a one-line change and no extra charge removes that choice. The attestation step matters as much as the isolation: it ties secret delivery to proof of what is running, not to which node a pod landed on.