Perimeter-Level Access Control (OpenClaw)

Description

OpenClaw's trust architecture: rather than evaluating safety per tool-invocation like Claude Code, OpenClaw places the trust boundary at the gateway perimeter using DM pairing codes, sender allowlists, and gateway authentication, with tool policy as configurable allow/deny lists per agent. Sandboxing (Docker/SSH/OpenShell backends) is opt-in and explicitly does not claim hostile multi-tenant isolation as a supported security boundary.

Key claims

Relations

Sources

src-20260423-0cff68d3291b