Palo Alto Unit 42 finds cascading MCP trust risk

The Model Context Protocol (MCP) has evolved through multiple security revisions since its 2004 launch to address significant authentication and authorization risks. Security testing by Palo Alto Networks' Unit 42 reveals that connecting five MCP servers to an AI agent produces a 78.3 percent attack success rate due to cascading trust vulnerabilities. Experts recommend using centralized identity-aware gateways, short-lived scoped tokens, and strict issuer validation rather than trusting individual servers. Open article →