Summary
A VentureBeat Pulse Research study across 116 enterprises reveals significant security vulnerabilities in AI agent deployments, with two-thirds enforcing scoped permissions but less than one in five isolating high-risk agents, leading to a majority experiencing security events or near-misses.