- Published on
Enterprises enforce agent permissions two-thirds of the time — and isolate high-risk agents less than one in five
- Authors
- Name
- DeepHorus
The vast majority of enterprises are failing to isolate their most dangerous AI agents, even as agentic AI adoption accelerates and security incidents mount, according to new VentureBeat Pulse Research. The survey of 116 enterprises with more than 100 employees found that 53% have agentic AI systems in production, with another 27% piloting them. That rapid deployment has already produced consequences: 53% of organizations report a confirmed agent security event or near-miss.
The central finding is a stark gap between monitoring and containment. While 65% of enterprises enforce scoped permissions at runtime and 56% monitor and log agent activity, only 18% isolate high-risk agents in sandboxes. Even among enterprises running agents in production, isolation is enforced just 21% of the time, and only 8% pair enforcement with isolation.
Identity management has improved but remains incomplete. About 49% of enterprises say each agent has its own scoped identity, but 63% report credential sharing somewhere in the fleet, and only 29% have a fully scoped fleet with no sharing. This persistence of shared credentials undermines the supposed per-agent identity controls.
The security stack is overwhelmingly borrowed from model providers and hyperscalers. OpenAI’s guardrails lead at 44%, followed by Microsoft Azure at 42%, Anthropic’s controls at 37%, and Google Cloud at 31%. A vast 92% of enterprises naming a primary security layer rely on a hyperscaler or model-provider-native solution.
Confidence is slipping. Thirty percent of enterprises now believe AI-armed attackers are ahead of their defenses, exactly matching the percentage that believe their defenses are ahead. The lack of isolation is the weakest layer precisely as autonomy scales, making containment the critical vulnerability in agentic security.
Security teams recognize that isolation limits blast radius, yet it remains the least enforced control. Despite satisfaction scores reaching a series high of 4.29 out of 5, churn intent is the highest recorded, with 74% planning to adopt, add, or replace agent security tooling within twelve months. This divergence suggests that enterprises are more determined than ever to overhaul their agentic security stacks, even as they express satisfaction with current tools.
