Agentic AI Exposes Hidden Enterprise Attack Paths

Josh Breaker-Rolfe reports that agentic AI creates a hidden exposure layer driven by two protocols—Model Context Protocol (MCP) and Agent-to-Agent (A2A)—which respectively expand access to external tools and spread context laterally between agents. The piece details four exposure patterns, including over-privileged access and context poisoning, and urges adoption of exposure management to map attack paths and prioritize mitigations.