How we secure 100+ clusters across 20+ clouds
Blog post from Baseten
Operating inference across more than 20 cloud providers and 100-plus clusters requires consistent security controls, especially because some providers offer less-hardened or less-uniform infrastructure. The company organizes its approach around four trust boundaries—infrastructure, operating partners, customer workloads, and internal access—and hardens each cluster before use through restricted network exposure, controlled access, patching, and data wiping at decommission. Customer code is treated as untrusted and isolated through microVM builds, separated image registries, tenant network controls, encrypted secrets, and runtime detection. To limit damage if a cluster or boundary is compromised, credentials are least-privileged and scoped to individual clusters, while supply-chain inputs such as configurations, images, model weights, and host software have dedicated safeguards. Centralized logs, alerts, perimeter scans, and provider reviews help verify that controls remain effective, and the post concludes with a checklist for auditing fleets or assessing infrastructure providers.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Kubernetes | 9 | 4,377 | 419 | 114 | -15% |
| Secrets Management | 4 | 2,876 | 539 | 148 | -7% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.