Autonomous Agents Threat Report Reveals alarming Statistics About Dormant AI Agents
New York, NY, USA, October 21st, 2026, CyberNewswire—A new report from Arrakis Security, titled the 2026 Autonomous Agents Threat Report, has raised significant concern among cybersecurity experts regarding the status of autonomous agents within organizations. According to the findings, a staggering 76% of active AI agents have not executed any actions in the past 60 days. In even more alarming terms, 44% of those agents have remained inactive for 180 days or longer, while their credentials and triggers still remain operational.
The research targeted a critical aspect of cyber safety: the prevalence of dormant agents that can still be activated from outside an organization. The report revealed that 24% of these inactive agents can be initiated through an unsigned webhook, posing a serious security risk. This leaves companies vulnerable, enabling potentially harmful exploitation by external entities.
The insights presented in the report were garnered from anonymized production telemetry collected over a 12-week duration, covering data from thousands of AI agents utilized by Arrakis Security’s clientele. Tal Baron, the company’s CEO and co-founder, shared his thoughts on the lack of formal data-driven discussions surrounding AI agent risks. “There’s a lot of discussion about AI agent risk, and very little data,” he remarked, highlighting a significant gap in organizational practices for managing agent-based access.
Baron underscored that when an employee departs from a company, their credentials are typically revoked immediately. However, the same diligence is rarely applied to AI agents. “When an agent stops being used, nobody removes anything,” he noted, emphasizing a concerning oversight in cybersecurity protocols—a reality in many organizations where ownership of these agents is often murky.
Other Findings:
The report also unveiled several additional findings, further illustrating the vulnerabilities inherent in autonomous agents:
-
22% of agents have the capability to send content to third-party models and act on the responses received. This opens the door to unauthorized or unintended data handling.
-
20% can pass tasks to other agents endowed with different permissions. This maneuvering allows them to obtain results they would not typically have access to, effectively bypassing established security protocols.
-
A significant 70% of agents can access sensitive systems, which include cloud storage, AI services, and marketing platforms. Alarmingly, almost none of these connections employ verified data loss prevention (DLP) techniques.
- Furthermore, 82% of agents possess at least one exposure to vulnerabilities, with 46% of them carrying multiple vulnerabilities. This multiplicity of risks poses an amplified threat landscape for organizations relying on these autonomous agents.
The full report, containing intricate details and analysis, can be accessed via the following link: Arrakis 2026 Threat Report.
About Arrakis Security:
Arrakis Security is dedicated to fortifying autonomous agents through a multi-faceted approach focusing on behavioral intelligence. This approach is anchored in understanding three critical layers: the agent’s identity, its actions, and the underlying motivations.
Their proprietary AI-SPM (Security Policy Manager) is instrumental in scrutinizing each agent’s structure, permissions, tools, and metadata on an hourly basis. This includes agents that remain idle, thereby revealing shadow agents, exposed keys, and various forms of misconfigurations.
Moreover, the company’s AI-TDR (Threat Detection and Response) module actively monitors actions in real-time, seeking deviations from pre-defined rules and policies. By modeling each agent based on its historical behavior, current posture, runtime, and context, it can assess how the agent reached its present stage and predict possible future actions. With each action taken by these agents, the system determines whether to allow, escalate, or intervene.
These findings underscore the pressing need for organizations to reassess their cybersecurity frameworks and address the mismanagement surrounding inactive or dormant AI agents. As organizations increasingly rely on autonomous systems, the importance of vigilant monitoring and effective management becomes paramount in safeguarding against potential vulnerabilities.