The Security Risks of AI Agents in the Enterprise | Kovrr

**Executive summary:** This report explains how agentic AI—autonomous agents that perceive, reason, plan, and act—introduces a distinct and expanding enterprise attack surface (indirect prompt injection, goal hijacking, persistent memory poisoning, over-permissioned credentials, unsafe function chaining, multi-agent cascades, and third-party integration risks) and recommends treating agents as a separate asset class with layered guardrails, quantified risk visibility, and continuous monitoring to prevent systemic failures and financial exposure.