SentinelAgent: Formal Framework for Verifiable Delegation Chains in Multi-Agent AI Systems

arXiv·high signal

Introduces SentinelAgent, a formal framework for verifiable delegation chains in federated multi-agent AI systems. When Agent A delegates to Agent B, which invokes Tool C on behalf of User X, no existing framework can trace the authorization chain or identify policy violations. The Delegation Chain Calculus (DCC) defines seven properties including authority narrowing, policy preservation, forensic reconstructibility, and cascade containment — six deterministic and one probabilistic.