You’ve probably lost sight of your network – here’s why (www.techradar.com)

As enterprises repatriate AI workloads from public clouds back to on‑prem data centers to regain control, privacy, and cost predictability for training, fine‑tuning and inference, they’re unintentionally creating hybrid environments that worsen network visibility and expand attack surface. Because many organizations keep some cloud services for data sourcing, collaboration, or burst scale, repatriation usually yields a fragmented stack where traffic crosses loosely integrated domains—exactly the blind spots attackers exploit. Kubernetes and hybrid tooling ease migration but also make IP‑centric controls brittle in dynamic, ephemeral environments. The antidote is a layered, fabric‑centric security model: strong IAM across environments, continuous endpoint and workload protection, automated cloud configuration hygiene, and cloud‑network security implemented as distributed firewalls embedded in the cloud fabric to enforce consistent policies and provide secure, high‑speed connectivity between data centers, clouds, and partners. Crucially, monitoring must correlate telemetry across identity, endpoint, posture and network to map attack paths and validate controls. Operationally, shift security left by baking guardrails into CI/CD, use cloud‑native constructs (CSP tags, Kubernetes namespaces) instead of IP rules, and adopt identity‑aware, distributed policy enforcement to reduce overhead and restore observability for AI/ML teams operating hybrid infrastructures.