Why 'monitoring' isn't enough for AI agents — and how I made delegation cryptographically verifiable

As AI agents gain autonomy, traditional logging and monitoring systems are proving insufficient for ensuring accountability. Developer Kironov Laziz-Del argues that server-side logs are merely claims, not proof, and proposes a shift toward cryptographically verifiable delegation. By using Ed25519 signatures, agents can sign their delegation instructions, allowing auditors to verify the chain of authority independently without trusting the server. The author emphasizes the importance of canonical serialization to prevent signature mismatches and suggests that capability-based security—where agents can only delegate a subset of their own permissions—is essential for safe multi-agent systems. The project, AI Control Tower, is available as an open-source tool to help developers implement these verifiable accountability measures. This approach addresses growing regulatory demands for transparency in autonomous systems, moving beyond simple audit trails to mathematical proof of authorization.