ray 2.58

Ray 2.58 introduces an experimental gVisor-backed sandboxing library developed with Google Cloud and Anyscale, enabling teams to execute model-generated code as isolated Ray-managed workloads. This update is relevant for enterprise AI teams using Ray for distributed training and inference, as it allows scheduling sandboxes, reserving CPU and memory, running commands, transferring files, and tearing down environments through the actor-based programming model. The release, announced in late August, addresses security concerns by preventing untrusted code from running directly in worker processes. For WindowsForum users, this highlights Ray's evolving role in secure AI workload management, though the feature is primarily aimed at cloud and Linux environments.