Study shows two-thirds of enterprises enforce AI agent permissions, few isolate risks

A recent study found that two-thirds of enterprises enforce scoped permissions for their AI agents at runtime, yet less than 20% effectively isolate their highest-risk agents. This research, conducted by VentureBeat across 116 enterprises, highlights significant vulnerabilities in agent security as businesses increasingly rely on autonomous systems.

The urgency for enhancing agent security arises from a growing number of incidents involving these systems. More than half of the surveyed enterprises have experienced a confirmed security event or a near-miss with their AI agents, raising concerns about the adequacy of current security measures. Despite efforts to manage permissions, credential sharing remains prevalent, affecting nearly two-thirds of agent fleets and further complicating the security landscape.

Confidence in agentic security is waning, as many enterprises now believe that AI-driven attackers are outpacing their defenses. This shift in perception comes amid a landscape where security strategies continue to rely heavily on tools borrowed from model providers and hyperscalers. The report suggests that these borrowed strategies may not be sufficient to address the unique challenges posed by autonomous agents.

Looking ahead, organizations must prioritize the isolation of high-risk agents and enhance their overall security posture. As the use of AI agents expands, the potential for security breaches could escalate, making it essential for enterprises to invest in more robust containment strategies. Failure to do so could not only lead to increased incidents but also a deeper erosion of trust in AI technologies.