Arrakis Security Report: 76% of AI Agents Remain Dormant but Retain Active Credentials
A new report by Arrakis Security reveals that a vast majority of AI agents are dormant but retain active credentials and triggers, posing significant security risks.
A comprehensive analysis by Arrakis Security has uncovered a critical security blind spot within organizations utilizing AI agents: a staggering 76% of these agents are dormant, meaning they have not been active for at least 60 days. Despite their inactivity, these agents retain their credentials and the ability to be triggered, presenting a substantial risk to enterprise security. The "2026 Autonomous Agents Threat Report" is based on anonymized production telemetry from thousands of AI agents across Arrakis customers' environments over a 12-week period.
Further exacerbating the risk, 44% of these dormant agents have been inactive for over 180 days, indicating a prolonged period where their potential for misuse has gone unaddressed. The report highlights that 24% of these inactive agents can be initiated by an unsigned webhook originating from outside the organization, creating an external attack vector that bypasses internal security controls.
"There's a lot of discussion about AI agent risk, and very little data," stated Tal Baron, CEO and co-founder of Arrakis Security. "When an employee leaves, their access is removed the same day. When an agent stops being used, nobody removes anything, because in most companies nobody owns it." This lack of ownership and oversight is a primary driver of the identified security gaps.
The study also revealed alarming capabilities within these AI agents. Approximately 22% of agents can interact with third-party models, sending data and acting upon the received answers, which could lead to sensitive information leakage or manipulation. Moreover, 20% of agents possess the ability to escalate tasks to other agents with different permission levels, potentially allowing them to access or modify data beyond their intended scope.
Compounding these issues, 70% of agents have access to sensitive systems, including cloud storage, AI services, and marketing platforms. Critically, almost none of these connections are protected by Data Loss Prevention (DLP) measures, leaving sensitive data vulnerable to unauthorized access, exfiltration, or modification.
The report quantifies the extent of exposure, with 82% of agents carrying at least one security exposure, and a significant 46% of those carrying more than one. This pervasive exposure underscores the urgent need for organizations to gain visibility and control over their AI agent deployments.
Arrakis Security's AI-SPM (Security Posture Management) solution scans agent structures, permissions, tools, and metadata hourly to surface shadow agents, exposed keys, and misconfigurations. Their AI-TDR (Threat Detection and Response) monitors actions in real-time, while the behavioral intelligence engine models each agent's unique history and context to identify and intervene in anomalous activities.
The findings serve as a stark warning to organizations rapidly adopting AI agents. Without proper lifecycle management, credential revocation, and continuous monitoring, these powerful tools can become significant liabilities, leaving sensitive data and systems exposed to exploitation.