Security Threat Modeling for AI-Agent Protocols (arXiv)

The first systematic comparative security analysis of MCP, A2A, Agora, and ANP. Key finding: identity forgery is a cross-protocol vulnerability — MCP relies on free-text names without cryptographic binding, A2A's JWT-based auth is vulnerable to agent card forgery, ANP's DID system lacks centralized revocation. Essential for choosing between agent communication protocols.