AI agents now pose security risks as companies deploy them without governance — enterprises must lock down non-human identities before attacks multiply

MIT Technology Review AI · April 21, 2026

AI Summary

•Companies are deploying AI agents (AI systems that make decisions and take actions independently) alongside human workers at scale, but many lack security controls to prevent these agents from being manipulated into accessing sensitive systems and stealing proprietary data.
•The problem is growing because non-human identities—digital permissions granted to software, bots, and now AI agents—are already outnumbering human user accounts at some large enterprises, and that gap will widen dramatically as agentic AI adoption accelerates.
•If you work in an enterprise, this means your company's data and systems are now exposed to a new type of attack: hackers compromising an unsecured AI agent to bypass firewalls and steal information. IT leaders and security teams must establish governance frameworks (policy + monitoring for AI agents) now, or face breaches that traditional human-focused security cannot prevent.

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.