77% of IT Managers Say Their AI Agents Are Out of Control - 5 Ways to Rein in Yours

Why It Matters

Uncontrolled AI agents threaten operational efficiency and expose enterprises to security breaches, making robust governance essential for sustainable AI adoption. Companies that establish clear guardrails now will avoid costly remediation and maintain trust in automated workflows.

Key Takeaways

• Only 23% of IT managers have full control over AI agents
• 81% say agents need more manual auditing than they save
• 86% anticipate agent proliferation will outpace security guardrails within a year
• Over half expect governance gaps within six months
• Effective governance requires traceability, intent, impact, success metrics, and rollback

Pulse Analysis

The Rubrik ZeroLabs survey highlights a growing governance crisis as AI agents proliferate across enterprises. While the technology promises rapid automation, 81% of IT leaders say the agents they deploy consume more time for manual checks than they save, eroding the promised productivity gains. This mismatch stems from the ease of spinning up agents—any team with API access can launch one in an afternoon—leading to a fragmented landscape reminiscent of early cloud adoption, where disparate frameworks and vendors created hidden security gaps.

Security experts warn that without a unified oversight model, agents become a moving target. The study shows 86% of respondents believe agent sprawl will outstrip existing guardrails within the next year, and more than half anticipate this breach occurring in six months. To counteract this, the report proposes a five‑question post‑deployment framework: trace what the agent did, understand why, catalog touched resources, assess success and cost, and pinpoint failures. Implementing these checks enables organizations to audit behavior, enforce policy, and retain the ability to roll back unintended actions—capabilities currently missing in most deployments.

Treating agent management as a first‑class discipline is no longer optional. Companies must separate orchestration, model, and governance layers, establish consistent identity models, and embed telemetry for continuous observability. By doing so, they can balance the speed of AI‑driven innovation with the rigor of security and compliance, ensuring that AI agents enhance rather than jeopardize business outcomes.