Trust

Trust and safety for hosted agents

ClawPilot hosts agent runtimes that can connect to tools, channels, credentials, and outside content. The useful trust stance is explicit boundaries, scoped permissions, and policies readers can inspect.

Agent risk boundaries

  • AI outputs can be wrong, incomplete, or unsafe to use without review.
  • Tool and channel access can turn a bad action into an external message or data operation.
  • High-risk decisions still need human oversight and the right workflow controls.

Control expectations

Use approvals, allowlists, access boundaries, and monitoring before live automations. Scope credentials and integrations to the workflow instead of granting broad access by default.

Openclaw, Hermes Agent, model providers, messaging platforms, and connected APIs keep their own documentation, policies, and failure modes. Hosting does not erase those dependencies.

Read the public policies

These pages state the current public boundaries more precisely than a generic security promise.

Privacy policyTermsAgent risk disclaimerContact ClawPilot