Concept代码沙箱 / 执行沙箱
Agent Sandbox
An isolated execution environment designed by OpenAI for agents to securely run code and make tool calls. Its architecture is built around three pillars—runtime security, persistence, and orchestration—supporting multi-layered security isolation from micro virtual machines to global orchestration.
Timeline (last 90 days)
Sep 7
Agent sandboxes are typically based on lightweight virtual machine (microVM) technology, such as Firecracker or Cloud Hypervisor, providing complete kernel isolation at the hardware virtualization level.
Unverified50%