MAGIQ: A Post-Quantum Multi-Agent AI Governance System with Provable Security

The researchers propose MAGIQ, a policy definition and enforcement framework for multi-agent AI systems that employs efficient post-quantum cryptographic protocols and provides provable security guarantees. The framework allows users to define communication and access control policy budgets for inter-agent sessions and tasks, including global budgets for one-to-many agent sessions, supports session-based policy enforcement, and enables agent accountability to users through message attribution. The authors formally model and prove the system's correctness and security under the Universal Composability (UC) framework, evaluate the framework's computational and communication overhead, and compare it with the existing agent AI framework SAGA. The text cites NIST recommendations that standard public-key algorithms such as RSA, Diffie-Hellman, and elliptic curve cryptography (ECC) be deprecated starting in 2030 and disallowed after 2035.