Over 90% of enterprises rely on cloud providers for AI agent security, yet 53% already faced incidents

Most enterprises deploying AI agents are unprepared to contain a rogue agent even after securing its credentials. Only 18% of companies isolate their highest-risk agents, and just 8% combine enforcement policies with isolation. Meanwhile, 53% of organizations have already experienced a security incident or near-miss related to agentic AI.

The July wave of VentureBeat Pulse Research found that 92% of enterprises naming a primary security layer default to their hyperscalers and AI platform providers. However, relying on provider-native controls for agentic security only widens the gap between what companies need and the measures they actually implement. Six waves of research, surveying 440 qualified enterprise security respondents since January, show this gap continues to grow.

A telling example comes from Visa: Rajat Taneja, the company's president of technology, explained at VB Transform 2026 how they used Anthropic's Mythos model to hunt for vulnerabilities in their own payment network. The model stitched minor weaknesses into working exploit chains, and Visa open-sourced the harness that governed the hunt. Yet such engineering depth is rare — most enterprises cannot act on what they find.

Satisfaction data does not match incident data. Last month, 46 enterprises reporting a confirmed incident or near-miss rated their satisfaction with security tooling at an average of 4.39 out of 5. Meanwhile, 30 of the 55 enterprises that experienced no incident rated their tools lower. This suggests companies tend to score familiar tools higher, even when those tools failed them or delivered mediocre results.

The research also revealed that 65% of enterprises enforce agent permissions at runtime, but few pair this with isolation. The agentic security market remains young, and the gap between what enterprises declare and what they actually deploy threatens their investments in agentic AI.