OpenAI Codex requires Linux kernel sandboxing

OpenAI's Codex agent runs exclusively on Linux, and the company has now spelled out exactly why: Windows lacks the kernel-level namespace isolation and seccomp syscall filtering that Linux provides natively, making it impossible to safely sandbox agentic code execution without layering in third-party hypervisors. The technical explanation matters most to enterprises running Windows-heavy infrastructure who want to self-host agentic coding systems. Without those primitives baked into the OS, any Windows-native deployment would depend on additional virtualization layers, introducing complexity, latency, and new attack surface. Linux's namespaces let Codex isolate each agent's process environment; seccomp filters limit which system calls that process can make. Windows has no direct equivalents at the kernel level. Essentially: (OpenAI, Microsoft) are on a collision course where Microsoft's own OS is structurally disadvantaged for hosting the agentic AI tooling that OpenAI -- its closest partner -- is shipping. - Linux namespace isolation and seccomp are prerequisites for Codex's sandboxing model, not optional hardening. - Windows Server environments cannot safely self-host Codex without hypervisor overhead that changes the deployment profile entirely. - Platform vendors, including Microsoft, face pressure to either build equivalent kernel primitives or accept that Linux dominates agentic AI infrastructure. As agentic code execution moves from research to enterprise standard, the Linux-Windows capability gap in kernel sandboxing will shape which organizations can self-host AI systems versus those forced into cloud-only deployments.