The MeshGuard
Agent Files

Named-author essays and technical guides on AI agent governance, API security, and delegated authority.

7 dispatches|1 categories

Does Constitutional AI Solve Enterprise Governance?

Behavioral principles can shape model choices. They do not replace identity, least privilege, approval, and runtime enforcement.

Behavioral principles can shape model choices. They do not replace identity, least privilege, approval, and runtime enforcement.

Model benchmarks are not deployment evidence. Agent systems also need tests for authority, side effects, delegation, replay, and fallback.

Authenticating a service is not the same as identifying an agent or proving its delegated authority for a specific action.

Content moderation evaluates what a model says. Production agent risk increasingly depends on what software is authorized to do.