AI Agent Security: Only 33% Have Least-Privilege Access

Only 33% of AI agents currently have least-privilege access, according to new research. This means the majority of AI agents have broader permissions than necessary. What's interesting is that 94% of organizations are confident their agents don't have too much access. However, two-thirds of these agents rely on standing permissions that are reviewed infrequently or not at all. The report also found that 65% of AI agents have taken actions outside their intended roles. For 29% of these, there was a measurable business impact. Another 36% were caught just before causing significant impact. The bottom line is that while many organizations have policies in place, there's a disconnect between written rules and actual enforcement for AI agent actions. This highlights a critical security gap that needs to be addressed as AI adoption grows.