54% of enterprises report AI agent security incidents
a venturebeat study of 107 firms found 54% faced ai agent security incidents as most still let agents share human credentials.
venturebeat surveyed 107 enterprises and found that 54% have already suffered a confirmed ai agent security incident or near-miss, revealing that autonomous tools are being deployed faster than the controls to contain them. the research highlights a critical identity weakness: only 32% of organizations provision unique, scoped identities for agents, while the majority allow agents to share human or service-account credentials. despite high satisfaction with native provider tools, only 30% of enterprises isolate high-risk agents, and a third fear their defenses are losing to ai-enabled attackers.