A Rubrik ZeroLabs survey reveals that only 23% of IT managers fell they have control over their AI agents, with many indicating that current security guardrails will soon prove insufficient.
The ease of creating AI agents has led to widespread proliferation, often beyond organizational oversight, resulting in inefficiencies where 81% of managers report agents demand more manual auditing than expected. Security concerns persist as unauthorized AI applications emerge both internally and from vendors. This expansion mirrors early cloud adoption challenges, where independent teams deploy agents using assorted frameworks, causing governance fragmentation and security vulnerabilities, as noted by Microsoft’s Kriti Faujdar. Most IT managers (86%) believe security measures will fail to keep pace with agent growth within the year, with over half forecasting this within six months.
Additionally, most lack the ability to reverse unintended agent actions, compounding risk. Experts warn that unregulated agent deployment leads to overlapping permissions and incomplete inventories, complicating management. Effective oversight requires enhanced telemetry to trace agent activities and enforce security at critical points. Post-deployment evaluation is essential but presently inadequate, encompassing questions about agent actions, motivations, interactions, success metrics, and failures. Without answers, organizations struggle to define acceptable behaviors, enforce access policies, and implement human intervention or rollback mechanisms. With autonomous agents posing greater risks than traditional software, there is a delicate balance between rapid deployment and governance. Successful organizations will prioritize agent management as a core discipline, adapting to evolving agent behaviors over time.
AI experts advise separating orchestration, model, and governance layers to maintain control, involving security, architecture, and business units rather than solely development teams, ensuring accountability and comprehensive oversight, but few organizations follow those recommendations.
The ainewsarticles.com article you just read is a brief synopsis; the original article can be found here: Read the Full Article…
