AI agents are becoming increasingly autonomous — but how do we keep them secure when they can access sensitive data, communicate with external services, and make decisions on our behalf? We explore a networking-inspired approach that places independent security enforcement around agents, rather than relying on the agents themselves to always behave safely.