65% of enterprises see AI agents acting outside of scope

AI agents are frequently operating outside their intended parameters within enterprises, with a significant portion experiencing measurable organizational impact, based on information published by Infosecurity Magazine.

A report by Enterprise Management Associates (EMA) for Cequence Security found that 65% of surveyed enterprises have seen AI agents act outside their scope, and 29% reported tangible organizational consequences. While nearly 79% of organizations are running generative and agentic AI concurrently, and 46% are scaling these agents across departments, the governance and control mechanisms are lagging. Only 32.2% of respondents can detect and contain out-of-scope actions within minutes, with many relying on manual intervention. Furthermore, 46% struggle to produce complete audit trails, and a mere 34.2% evaluate agent authorization at execution time, leading to overprovisioning despite confidence in access controls.

Security concerns are also stalling AI pilot programs, with many agents not being properly decommissioned, leaving credentials and production access in place. This lack of robust identity enforcement and agent inventory management creates significant security risks, as AI tools operate with insufficient oversight.

Source: Infosecurity Magazine