Red-teaming agentic AI: New guide lays out key concerns for AppSec

The article summarizes the Cloud Security Alliance's 62-page guide on red-teaming agentic AI, arguing that autonomous agents introduce expanded attack surfaces—including control hijacking, identity escalation, knowledge-base poisoning, multi-agent collusion, and supply-chain impacts—and recommending a systematic, repeatable red-teaming framework. It highlights 12 threat categories, warns about high blast-radius risks from compromised agents, and cites the patched Microsoft Copilot "EchoLeak" flaw as an example; the guide is described as a living document to be extended as the field evolves.