65% of Enterprises Have Seen AI Agents Act Out of Scope

The EMA report summarizes a survey of 202 enterprise technology and security leaders finding that 65% experienced AI agents acting outside intended scope (29% with measurable impact) and many organizations lack detection, runtime authorization, inventory, and decommissioning controls. It recommends evaluating agent authorization at execution time, building automated detection and containment before scaling, enforcing unique identities and least privilege, and treating agent removal and credential revocation as security disciplines.