.svg){kind=spacer}
May 14, 2026
Enterprise adoption of agentic AI is advancing quickly in pilots but stalling before true production deployment because organizations do not yet trust the security controls around autonomous systems. Based on Cisco’s survey of senior IT and security leaders, 85% of organizations are already experimenting with, piloting, or deploying agentic AI in some form, yet only 5% have reached broad production use. The post says this gap is driven mainly by security concerns: nearly 60% of security leaders view security as the top barrier to wider adoption, and the three biggest worries are agent access control, data exfiltration, and agent autonomy and behavior. Cisco presents these not as edge cases but as structural risks, since autonomous agents can take actions, not just access information, and therefore introduce a wider operational threat surface.
The article also explains that the most successful deployments so far are largely internal-facing, in areas such as IT operations, SecOps, internal finance, and R&D, while customer-facing autonomous agents remain mostly stuck in pilot because the reputational and security risks are higher. Cisco argues that traditional identity and access management is no longer enough for this environment and that Zero Trust must evolve from a model centered on who an entity is to one focused on what it is allowed to do. The company also highlights fragmented accountability as a major obstacle: 29% of respondents say the CISO owns agentic AI security, 27% say the CIO or IT team, 24% point to a central AI committee, and 11% say there is no clear owner at all. The post’s conclusion is that organizations will only move agentic AI safely into production if they define non-human identities clearly, apply least-privilege and behavior controls, and establish unified oversight so autonomous systems operate within tightly observable boundaries.
