CloudSEK Links Silent Lynx to AI-Assisted Tajik Document Theft

Researchers uncovered the theft after finding stolen Tajik government files and AI activity logs in a public folder on a server linked to Silent Lynx.

(For more news, click here)

Government document systems connect agencies that handle sensitive correspondence. Their security depends on controlling which records each account can reach.

CloudSEK says the Silent Lynx espionage group collected more than 2,000 defense-related documents from Tajikistan’s government document platform. Its investigation also describes persistent access to the country’s railway network. The attackers used an AI agent to support reconnaissance, document collection and cleanup.

CloudSEK found an internet-accessible folder on a server it attributed to Silent Lynx. No login was required to view its contents. The folder held stolen documents, compromised credentials, malware and records of the operator’s conversations with an AI agent. These files allowed researchers to reconstruct activity spanning April to October 2026.

One account opened access across agencies

The attackers entered Tajikistan’s national document platform using a railway employee’s standard account, according to CloudSEK. The platform lets ministries and state bodies exchange official correspondence. Weak checks on individual documents let the attackers retrieve records beyond the railway. A separate file service provided attachments without requiring authentication.

The collected material included personnel transfers, flight-clearance requests, defense correspondence and identity documents. The exposure shows how a single compromised account could reach sensitive records held across the government systems.

The collection also contained material associated with the National Guard and the State Committee for National Security. CloudSEK found hundreds of State Investment Committee letters as well. This extended the exposure beyond defense correspondence to other government business.

CloudSEK also found domain access from mid-August until at least October 10. The attackers obtained administrative access and used it to copy operational reports and internal correspondence. There was no reported disruption of railway services.

AI supported collection and cleanup

The operator used Hermes, an open-source AI agent framework, primarily with DeepSeek models. CloudSEK found instructions, conversation records and agent memory alongside the stolen data. These records showed how the operator delegated work and continued tasks across sessions.

CloudSEK told Reuters:

“Hermes planned the attacks, wrote the rules, ran them, kept notes, and wrote up reports.”

Reuters also reported that the agent wrote software to download, organize and divide the files for transfer. CloudSEK found collectors designed to resume after interruptions and send progress updates to Telegram. This helped sustain collection across sessions.

CloudSEK’s investigation describes cleanup of histories and logs. The exposed directory survived those efforts. The configuration permitted some destructive commands without further approval.

The link to Silent Lynx

Silent Lynx was already known for phishing attacks against Central Asian government institutions. Seqrite documented campaigns against Kyrgyzstan’s financial institutions in January 2025. CloudSEK’s investigation shows how an operator linked to the group used AI to support the theft of Tajik government records.

CloudSEK attributed the exposed server to Silent Lynx with high confidence. It matched malware code markers and build paths with those in previously documented Silent Lynx tools. Their assessment of a Kazakhstan-based nexus remained at medium confidence. Residential internet connections, Russian-language instructions and activity patterns supported that assessment.

The findings rest on CloudSEK’s analysis of attacker-side records. Reuters reported that Kazakh and Tajik officials had not immediately responded to its requests for comment.

The case points to AI’s role in sustaining espionage after an attacker gains access. Weak access controls can expose sensitive records across agencies. An automated collector can make that exposure easier to exploit at scale.