The report
A compromised account and a surprisingly permissive government-document service can make the difference between a narrow intrusion and large-scale collection. In a Reuters account of CloudSEK's investigation, researchers described an exposed server associated with the Silent Lynx cluster, where an operator used a Hermes-based AI workflow while collecting documents accessible through Tajik government systems.
What was reported, and what was not established
The investigators said access involving a railway employee's account exposed cross-agency document routes and attachments. More than 2,000 defence-related documents were reportedly retrieved or available for collection, including correspondence routed through security and government bodies. That description is not proof that every document was classified or that an AI model independently selected every target and executed each collection step. It is evidence of reported operator-directed use of an agent among other intrusion tools. See the Silent Lynx Case.
The exposed operator infrastructure reportedly contained conversation history, tools and supporting records. Such traces matter because they can sometimes distinguish instructions from the human operator, suggested actions from a model and actual tool effects. A shared workspace does not show that every compromise attributed to the cluster was performed by the same agent, much less by an entirely autonomous system.
Evidence limits
This Compose pass verified accessible syndicated Reuters coverage but could not directly retrieve the full original CloudSEK write-up. It therefore does not independently certify the detailed file-by-file collection timeline, a victim authority's confirmation, or a definitive sponsor. Continued work should compare agent transcripts to access logs, downloaded objects and retained authentication records before assigning the agent a larger role or estimating the sensitive information actually lost.