Four of five enterprises that secured AI agent identities still can't contain one that goes rogue

VentureBeat Pulse research across 440 enterprise security respondents found that 53% of enterprises have had an AI agent security incident or near-miss. While 65% enforce agent permissions at runtime, only 18% isolate high-risk agents and just 8% combine enforcement with isolation, leaving major gaps in agentic security.

The data suggests many enterprises are overconfident in familiar hyperscaler and AI platform controls, even when those tools do not prevent incidents. As AI agents gain more autonomy, containment is becoming a distinct security requirement rather than a feature that can be delegated to existing cloud controls.