Agentic AI is not Secure

AuthZed argues that current AI agent communication protocols including MCP, A2A, and ACP lack adequate authorization, exposing systems to vulnerabilities such as insufficient token granularity, inadequate revocation propagation, and unclear consent flows.