Agentic security gaps persist as enterprises deploy AI agents in production
venturebeat.com

Agentic security gaps persist as enterprises deploy AI agents in production

Tech News
2 min read

Published by AINave Editorial • Reviewed by Ramit

TL;DRA VentureBeat survey of 116 enterprises reveals a containment gap: most enforce runtime permissions and monitor agents, but fewer than one in five isolate high-risk agents, even as 53% report security events.
A new VentureBeat Pulse Research survey of 116 enterprises reveals a critical gap in agentic security: most organizations enforce runtime permissions and monitor agent activity, but fewer than one in five isolate high-risk agents. With 53% of enterprises already running agentic AI in production and 53% reporting a confirmed security event or near-miss, the missing containment layer is the weakest link as agent autonomy scales.\n\n## Agent deployments are live, but containment is missing\n\nMore than half of enterprises (53%) have agentic AI systems in production today, and another 27% are piloting or running a limited rollout. The security exposure has scaled with deployment: 53% of organizations have already had an agent security event, with 19% confirming an incident and 38% identifying a near-miss caught before harm. Among enterprises describing their security posture, 65% enforce scoped permissions at runtime and 56% monitor and log agent activity, yet only 18% isolate high-risk agents in sandboxes. Even among enterprises running agents in production, isolation reaches just 21%, and only 8% pair enforcement with isolation. That ordering is backwards from a defense-in-depth standpoint: observation tells you what happened, enforcement tries to prevent it, but isolation limits the blast radius when prevention fails.\n\n## Identity is improving, but credential sharing still widens blast radius\n\nPer-agent identity is now the most-cited pattern: [

Sources

Latest Tech News