ARRAKIS Security’s 2026 Autonomous Agents Threat Report analysed anonymised telemetry from thousands of AI agents across Arrakis customers over a 12-week period. The findings show that 76% of active agents have not run in 60 days, yet their credentials remain usable and their triggers still fire; 44% have not run in 180 days. A further 24% of dormant agents can be started by an unsigned webhook from outside the organisation.
The study emphasises that when an employee leaves, access is removed promptly, but when an agent simply stops being used, there is often no ownership or removal of the agent.
Additional highlighted risks include that 22% of agents can send content to a model run by a third party and act on the response, while 20% can hand off tasks to another agent with different permissions and obtain results they are not authorised to produce. The report also notes that 70% of agents can reach sensitive systems such as cloud storage, AI services and marketing platforms, with most connections lacking verified data loss prevention.
Overall, 82% of agents carry at least one exposure, and 46% of those carry more than one. The full report is available at arrakis[.]security/2026-threat-report. The findings underscore the need for clearer ownership, continuous monitoring, and stronger controls around dormant agents and cross-agent interactions.