AI agent governance is emerging as a critical enterprise challenge as organizations struggle to manage the rapid growth of autonomous systems. According to a survey by Rubrik ZeroLabs, only 23% of IT managers report having complete control over their AI agents, while 77% admit their agents are effectively “out of control.” This highlights a growing risk as AI adoption accelerates faster than governance frameworks can keep up.
The core issue is agent sprawl. AI agents are easy to deploy, often requiring minimal technical effort, which leads to widespread unsanctioned usage. Employees and teams can quickly create agents using APIs or third-party tools, bypassing traditional security controls. This mirrors the early days of cloud adoption, where decentralized deployments created fragmented systems and hidden risks.
The consequences are significant. Around 81% of IT managers say AI agents require more manual auditing and monitoring than expected, which reduces the productivity gains that the agents were designed to deliver. At the same time, 86% expect agent proliferation to outpace security guardrails within the next year, with more than half anticipating these developments within six months.
- Only 23% of organizations have full control over AI agents.
- 81% report increased manual oversight instead of efficiency gains.
- Agent sprawl creates security and governance gaps.
- Most enterprises lack rollback and audit capabilities.
A major challenge is visibility. Many organizations cannot answer fundamental questions about their agents, such as what actions they took, why they made decisions, or what data they accessed. Without traceability, it becomes difficult to audit behavior, enforce policies, or recover from failures.
To address these issues, enterprises must treat AI agent governance as a first-class discipline. This includes establishing clear policies, implementing observability and audit trails, and ensuring human oversight for critical decisions. Separating orchestration, model, and governance layers can also help maintain control and accountability.
Ultimately, AI agent governance will determine whether organizations can scale automation safely. Those that prioritize control, transparency, and security will be better positioned to unlock the benefits of agentic AI without exposing themselves to unmanaged risk.
ソース：
https://www.zdnet.com/article/it-managers-say-ai-agents-are-out-of-control/
