AI Agents Are Getting Out of Control, Warns IT Leaders

 

A growing number of IT managers are raising concerns over the rapid spread of artificial intelligence (AI) agents in the workplace, with new research showing that many organizations are struggling to keep them under control.

According to a survey by Rubrik ZeroLabs, only 23% of IT managers say they have full control over the AI agents operating within their systems. Meanwhile, 77% admit that these tools are becoming difficult to manage, as more employees and teams deploy them independently.

The report highlights a major issue: AI agents are easy to create but hard to govern. Many users bypass security measures—such as VPNs—to quickly launch AI assistants, resulting in a surge of unsanctioned applications across organizations. This “agent sprawl” is now being compared to the early days of cloud computing, when systems expanded faster than companies could regulate them.

Even more concerning, 81% of respondents said AI agents require more time for monitoring and auditing than they actually save in productivity. Security risks are also rising, with most IT leaders warning that existing safeguards may not keep up with the pace of AI adoption in the coming months.

Experts from companies like Microsoft emphasize that without proper governance, organizations risk building systems that are difficult to trust, scale, or secure. The lack of visibility into what AI agents are doing—such as what data they access or why they make certain decisions—adds to the challenge.

To address this, the report suggests that companies treat AI agent management as a “first-class discipline.” This includes implementing stronger oversight, clearer policies, and better tracking systems to monitor agent behavior and prevent unintended actions.

As AI continues to evolve, businesses face a critical balancing act: moving fast to stay competitive while ensuring their systems remain secure and under control. Without proper safeguards, the very tools designed to boost productivity could end up creating new risks instead.