#sandbox-security

Boundary-Bench measures how much coding agents lose when you lock down the sandbox

A new open-source benchmark runs coding agents in hardened, enterprise-style sandboxes to quantify the capability cost of security controls.