AI Agents: The Underestimated Insider Threat
AI agents have long been an integral part of the digital workforce—yet many companies still underestimate the security risk they pose. Erich Kron, CISO Advisor at Knowbe4, explains why autonomous AI systems are becoming the new insider threat and what companies need to do to address it.
When companies talk about digital identities, they no longer mean just their human employees, their service accounts, automation platforms, and bots. That’s because, just under two years ago, a new and entirely different type of digital identity emerged: AI agents. Unlike traditional machine identities, AI agents do not follow a rigid sequence of commands. They operate independently and autonomously. They can plan, select tools, call APIs, generate code, and even delegate tasks.
Significant potential for damage—barely recognized
The wide range of their applications, their autonomy, and their often very extensive access rights make AI agents a source of risk with significant potential for damage. They pose a security risk whose true extent most companies are only now beginning to grasp. According to a study by the Cloud Security Alliance, 82 percent of all companies currently use AI agents without their cybersecurity teams even being aware of their existence. According to a recent CyberArk study, only 32 percent of all companies have implemented controls to manage their AI-related risks. And according to a Keyfactor study, only 28 percent believe they can prevent an AI agent from causing damage within their ecosystem.
It’s no wonder that nearly two-thirds of all companies surveyed experienced at least one security incident last year related to AI agents. The consequences were sometimes significant: 61 percent of the incidents led to data breaches, 43 percent to operational disruptions, and 35 percent even to financial losses.
New attack vectors are emerging rapidly
A key cause of these security incidents is the compromise of AI agents by attackers. Worldwide, prompt injection, goal hijacking, privilege escalation, and memory poisoning have quickly become popular attack vectors. It is difficult for a company’s cybersecurity team to detect and defend against such attacks in a timely manner. The attacker does not need to steal login credentials. It is entirely sufficient for the attacker to manipulate the inputs that assign tasks to the AI agent—for example, through malicious emails, support tickets, or websites.
To make matters worse, the API keys for AI agents are usually valid indefinitely, their access permissions are far too broad, and oversight of them is generally insufficient. As a result, AI agents provide the ideal conditions to become the next insider threat.
Transparency as the First Step Toward Security
To successfully manage this risk, companies must first and foremost do one thing: create transparency. An inventory of all AI agents used in the IT ecosystem is an absolute necessity. Then, the Zero Trust principle must be consistently applied to all AI agents. An agent used for ticket summarization must not be granted access to confidential customer databases. Furthermore, permanent access rights should be replaced with temporary ones.
To detect unusual AI activity as early as possible, security teams should shift their focus from verifying an agent’s identity to monitoring its behavior. Solutions—such as Agent Risk Manager—that can effectively support cybersecurity teams in taking measures to minimize the risks posed by their AI agents are already available on the market.
Cybersecurity professionals must recognize that AI agents have long been an integral part of their company’s digital workforce. They require different security measures than human and machine identities. They can easily be manipulated and turned into an insider threat. It is essential to protect against this—with sound policies and comprehensive digital workforce security that helps safeguard all parts of the new digital workforce.
More information: www.knowbe4.com
This article originally appeared on m-q.ch - https://www.m-q.ch/de/ki-agenten-die-unterschaetzte-insider-bedrohung/
