On April 21, 2026, the Cloud Security Alliance (CSA) in collaboration with Token Security released a comprehensive research report titled The State of AI Agent Security. The study reveals that 66% of surveyed organizations have experienced at least one cybersecurity incident directly attributed to the deployment of autonomous AI agents. These agents, designed to perform tasks independently across cloud environments, have introduced a new layer of risk that many enterprise security frameworks are currently unequipped to handle.
The report focuses on the proliferation of non-human identities (NHIs), which include AI agents, service accounts, and automated bots. According to the data, the average enterprise now manages over 10,000 unique AI-driven identities, outnumbering human users by a ratio of 45 to 1. Technical analysis within the report indicates that 72% of these agents are granted over-privileged access, meaning they possess permissions far exceeding what is required for their specific functions. This excessive access was cited as the primary vector in 58% of the reported incidents, leading to unauthorized data exfiltration and lateral movement within corporate networks.
A significant finding of the research is the lack of lifecycle management for AI agents. Approximately 81% of organizations admitted they do not have a formal strategy for decommissioning AI agents once their specific task or project is completed. These abandoned identities, often referred to as zombie agents, remain active in the background, maintaining access to sensitive databases and APIs. The CSA notes that these identities are frequently targeted by attackers because they are often excluded from standard multi-factor authentication and monitoring protocols.
Hillary Baron, Senior Technical Director at the CSA, stated that the speed of AI adoption has outpaced the development of governance frameworks. The report highlights that while 90% of firms are increasing their use of AI agents for operational efficiency, only 24% have integrated these agents into their existing Identity and Access Management systems. Token Security CEO Itay Rozenman emphasized that the shift from human-centric to machine-centric security is the most significant challenge facing Chief Information Security Officers in 2026. The report concludes by recommending the implementation of automated discovery tools and the enforcement of least-privilege principles for all non-human entities.
The survey included responses from 1,500 IT and security professionals across North America, Europe, and Asia-Pacific, representing sectors including finance, healthcare, and manufacturing. The findings underscore a systemic vulnerability in the agentic workflow model, where AI systems interact with other software components without human intervention. As of April 2026, the CSA is calling for a standardized industry framework to govern the creation, monitoring, and retirement of AI agents to mitigate these rising security threats.