Arrakis Security report: 76% of active AI agents dormant for 60+ days yet retain working credentials and triggers
The 2026 Autonomous Agents Threat Report from Arrakis Security, based on 12 weeks of anonymized production telemetry from thousands of AI agents in its customers' environments, finds that 76% of active agents have not run in 60 days, while their credentials still work and their triggers still fire. 44% have not run in 180 days. According to Tal Baron, CEO and co-founder, nobody removes access for unused agents because in most companies nobody owns them. The report is vendor-published research and the figures have not been independently verified.
Detalhes técnicos
- 24% of dormant agents can be started by an unsigned webhook from outside the organization.
- 22% of agents can send content to a third-party-run model and act on the answer.
- 20% can pass a task to another agent with different permissions and receive results they are not allowed to produce themselves.
- 70% can reach sensitive systems (cloud storage, AI services, marketing platforms), and almost none of those connections have verified DLP.
- 82% of agents carry at least one exposure, and 46% of those carry more than one.
Remediação
The article gives no explicit guidance. Its implied recommendation is to assign ownership of AI agents and to revoke access for unused agents, as is done when an employee leaves. The vendor promotes its own AI-SPM (hourly posture scanning, including agents that never run) and AI-TDR (real-time behavioral monitoring) products.
Linha do tempo
-
Arrakis Security report: 76% of active AI agents dormant for 60+ days yet retain working credentials and triggers
ler a análise desta fonte ocultar
The 2026 Autonomous Agents Threat Report from Arrakis Security, based on 12 weeks of anonymized production telemetry from thousands of AI agents in its customers' environments, finds that 76% of active agents have not run in 60 days, while their credentials still work and their triggers still fire. 44% have not run in 180 days. According to Tal Baron, CEO and co-founder, nobody removes access for unused agents because in most companies nobody owns them. The report is vendor-published research and the figures have not been independently verified.
Technical Details
- 24% of dormant agents can be started by an unsigned webhook from outside the organization.
- 22% of agents can send content to a third-party-run model and act on the answer.
- 20% can pass a task to another agent with different permissions and receive results they are not allowed to produce themselves.
- 70% can reach sensitive systems (cloud storage, AI services, marketing platforms), and almost none of those connections have verified DLP.
- 82% of agents carry at least one exposure, and 46% of those carry more than one.
Remediation
The article gives no explicit guidance. Its implied recommendation is to assign ownership of AI agents and to revoke access for unused agents, as is done when an employee leaves. The vendor promotes its own AI-SPM (hourly posture scanning, including agents that never run) and AI-TDR (real-time behavioral monitoring) products.