The Growing Challenge of AI Agent and NHI Management

The article outlines growing security risks posed by agentic AI and non-human identities (NHIs), describing attack vectors such as prompt injection, jailbreaking, data poisoning, malicious proxy uploads that can exfiltrate API keys and user data, and a referenced zero-click vulnerability (CVE-2025-32711) against Microsoft Copilot. It emphasizes the rapid scale of NHIs, the likelihood of future adversarial attacks on agents, and provides ten practical mitigation steps (fingerprinting agents, lifecycle ownership, permissions management, DLP, sandboxing, input sanitization, etc.) to reduce risk.