--- url: "https://www.xcademia.com/news/exposed-silent-lynx-server-reveals-ai-assisted-cyber-espionage-against-central-asian-governments" title: "Exposed Silent Lynx Server Reveals AI-Assisted Cyber Espionage Against Central Asian Governments" description: "CloudSEK uncovers Silent Lynx cyber espionage targeting Central Asian governments, exposing stolen defence documents and AI-assisted operations." publishedAt: "2026-11-12T08:36:10.911+00:00" updatedAt: "2026-11-12T11:45:59.556634+00:00" type: news category: cybersecurity source_name: CloudSEK source_url: "https://www.cloudsek.com/blog/off-the-rails-inside-an-exposed-silent-lynx-server?" tags: - "#SilentLynx" - "#CyberEspionage" - "#AIinCybersecurity" - "#DeepSeek" - "#APT" - "#ThreatIntelligence" - "#CriticalInfrastructure" - "#GovernmentCyberSecurity" --- # Exposed Silent Lynx Server Reveals AI-Assisted Cyber Espionage Against Central Asian Governments > An exposed server has uncovered Silent Lynx cyber espionage operations targeting Central Asian governments and critical infrastructure, revealing stolen defence documents, compromised railway networks and AI-assisted attack operations. Source: **CloudSEK** · 12 November 2026 An exposed server operated by the Silent Lynx cyber espionage group has provided security researchers with an unusually detailed view of its activities targeting government organisations and critical infrastructure across Central Asia. In an investigation published on 11 November 2026, cybersecurity company CloudSEK revealed that an internet-accessible server contained malware, stolen government documents, compromised credentials, intrusion records and operational data associated with the group's activities. The investigation identified compromises involving Tajikistan's national railway network, government document management infrastructure and Uzbekistan's Smart Forestry platform. It also uncovered phishing materials prepared to impersonate Turkmenistan government organisations. One of the most significant findings was the presence of a self-hosted artificial intelligence agent powered primarily by DeepSeek models. The agent supported reconnaissance, information collection, task coordination and evidence cleanup, providing insight into how AI-assisted workflows are being incorporated into cyber espionage operations. CloudSEK assessed with high confidence that the exposed infrastructure belonged to a Silent Lynx operator. The investigation also identified technical and network-level indicators supporting a possible Kazakhstan nexus. The findings highlight the security implications of compromised government accounts, weaknesses in interconnected public-sector systems and the growing operational use of AI by threat actors. ## What Is Silent Lynx? Silent Lynx is a cyber espionage cluster that has been publicly tracked since January 2025. Security researchers have associated the group with targeted phishing campaigns against government organisations in Central Asia and neighbouring regions. Previous investigations documented the use of malicious Windows shortcut files, archive-based delivery techniques and lightweight reverse shells to establish access to targeted systems. Different security vendors have tracked related activity under several names, including UAT-6275, TAG-193, Cavalry Werewolf and ShadowSilk. These names reflect overlapping research and attribution assessments rather than necessarily representing identical operational groups. Earlier reporting connected Silent Lynx activity to government organisations, diplomatic institutions and critical infrastructure. CloudSEK's investigation adds another layer to this understanding by exposing the group's internal working environment, including its malware development files, victim information, infrastructure records and AI agent conversations. ## Inside the Exposed Silent Lynx Server The investigation began with an open directory hosted on a Debian virtual private server associated with Baxet Group infrastructure. The server was accessible without authentication through port 8088. It contained more than a dozen gigabytes of data distributed across hundreds of thousands of files. Rather than separating development tools, operational infrastructure and collected information, the operator had stored these materials within the same environment. The exposed directory contained several important components: - **fish/:** Phishing materials, malicious Windows shortcuts, PowerShell payloads, decoy documents and callback logs. - **payloads/:** Malware builds organised according to target countries and victims. - **dikii/:** Individual working directories containing reconnaissance results, credentials and collected documents. - **.c2/:** Webshell-related files and captured output from the Uzbekistan Smart Forestry intrusion. - **.hermes/:** AI agent configuration, memory, scripts, task records and conversation logs. - **.ligolo-mp-server/ and .nxc/:** Network pivoting sessions and SMB credential databases. - **tools/ and CVE directories:** Offensive security utilities, network discovery tools and publicly available exploit resources. The exposure provided researchers with a timestamped record of activity spanning April to October 2026. It also revealed how the operator organised different stages of an intrusion, from initial delivery and reconnaissance to credential collection, lateral movement and data extraction. ## How Silent Lynx Delivered Its Malware CloudSEK identified a phishing delivery chain built around malicious Windows shortcut files disguised as ordinary PDF documents. The operator used ZIP archives, password-protected ZIP files and ISO images to distribute these shortcuts. The approach combined a convincing document-themed lure with hidden PowerShell execution. ### The Infection Process The investigation documented the following sequence: 1. **Initial delivery:** A target receives an archive or ISO containing a Windows shortcut displaying a PDF-style icon. 2. **Shortcut execution:** Opening the shortcut displays a decoy document while initiating a hidden PowerShell process. 3. **Payload retrieval:** PowerShell downloads a script named P.ps1 from the operator's server. 4. **Remote shell establishment:** The script opens another decoy PDF and establishes a raw TCP connection to the attacker's infrastructure. 5. **Initial reconnaissance:** A dispatcher script automatically executes commands such as whoami, hostname and ipconfig to collect basic information about the compromised system. The collected information helps the operator understand the environment and identify potential next steps. CloudSEK also identified an additional delivery variant designed to bypass Microsoft's Antimalware Scan Interface (AMSI). The variant modified the AMSI initialisation state through reflection before downloading the PowerShell payload. According to the investigation, this technique was not documented in previous public reporting on Silent Lynx. The discovery suggests that the group was developing and testing multiple delivery approaches rather than relying on a single phishing method. ![info-1](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1794472436922-info1--161-.webp) ## Turkmenistan Emerges as a Potential Target The exposed server also contained phishing materials prepared to impersonate Turkmenistan organisations. One decoy document imitated an electronic railway ticket for a journey between Ashgabat and Turkmenabat in October 2026. Another executable contained a long Turkmen-language filename referencing a report associated with the Ahal regional administration's 2026 state programme. CloudSEK found that the executable contained references to the operator's infrastructure and the administration's domain. The operator also tested the executable on their own machine. These findings indicate preparations for a potential campaign targeting Turkmenistan. However, the presence of test materials does not independently establish that the intended campaign was successfully executed. The investigation also noted earlier Silent Lynx activity involving Turkmenistan-related organisations and lures. ## Three Major Intrusions Revealed by the Investigation The most significant findings concern three separate compromises involving Tajikistan and Uzbekistan. ### 1. Tajikistan Railway Network Compromise CloudSEK identified evidence of persistent access to Tajikistan's national railway network. The operator maintained domain-level access from mid-August until at least 10 October 2026. A network tunnelling database showed an active connection through a railway host for nearly two months. After establishing access, the operator conducted internal reconnaissance and collected credentials and operational information. The investigation identified several activities: - Enumeration of dozens of internal Windows systems, including Windows 7 and Windows Server 2022 environments. - Administrative access to multiple servers using plaintext passwords and NTLM hashes. - Creation of NTDS-format credential dumps containing accounts such as Administrator and krbtgt. - Collection of railway operational reports, dispatcher summaries and internal correspondence. - Access to credentials associated with upstream ISP network equipment, including MikroTik, Cisco and D-Link devices. CloudSEK also found an ASP.NET webshell deployed on a railway web server. The operator had prepared Metasploit resource files targeting MS17-010 against an unpatched Windows Server 2008 R2 system. The investigation therefore points to a compromise extending beyond a single workstation, involving access to internal infrastructure, administrative credentials and network-related resources. ### 2. Tajikistan Government Document Management Platform The second intrusion exposed a significant weakness in the country's national electronic document management platform. The platform is used by government ministries and state organisations to exchange official correspondence. According to CloudSEK, the operator accessed the platform through its government single sign-on portal using credentials belonging to a railway employee with a standard user account. The account's permissions were not sufficient to explain the breadth of information accessible through the platform. CloudSEK identified weak per-document authorisation checks and an attachment service that returned files without authentication. The operator developed automated collection scripts capable of processing large numbers of documents, resuming interrupted operations and reporting progress through Telegram. The collected information included: - Thousands of documents addressed to or routed through the Ministry of Defence. - Correspondence involving the National Guard. - Documents associated with the State Committee for National Security. - Hundreds of State Investment Committee letters. - Government organisation directories containing department and user information. The collection included more than 2,000 unique documents associated with the Ministry of Defence. Their subject matter covered military personnel transfers, defence procurement, airspace permissions, military attaché assignments, foreign liaison activities, budgets, training exercises and air-defence cooperation. Some collected files also contained passport and national identity document scans associated with official visits and delegations. This incident demonstrates how a compromised employee account can become a pathway to information beyond the employee's immediate organisational responsibilities when application-level authorisation controls are inadequate. ### 3. Uzbekistan Smart Forestry Platform Compromise CloudSEK also reconstructed the intrusion into Uzbekistan's government-run Smart Forestry platform. The platform uses GeoServer technology for geospatial information. The investigation revealed an attack chain that progressed from a web application compromise to database-level access and command execution on the underlying host. The operator reportedly: 1. Deployed malicious WAR files to the GeoServer instance, including a file containing a Metasploit payload servlet. 2. Established a webshell through the application's /health/ path. 3. Accessed PostgreSQL using superuser privileges. 4. Used PostgreSQL's COPY FROM PROGRAM functionality to establish a reverse shell. 5. Created scheduled cron entries to maintain callbacks to external infrastructure. The investigation also identified limitations in the operator's access. CloudSEK stated that the attacker did not obtain root privileges. Attempts to use sudo required a password, while access to the Docker socket was denied. Nevertheless, the compromised database and host-level command execution provided substantial control over the forestry platform's database and associated application data. The findings also add technical detail to activity previously associated with the Smart Forestry initiative in May 2026. ![info-2](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1794472470391-info2--162-.webp) ## The AI Agent Behind Silent Lynx Operations One of the most notable discoveries was the presence of an AI-powered operational assistant installed directly on the exposed server. The operator used Hermes, an open-source autonomous agent framework, to coordinate different activities. The installation contained persistent memory, task management functionality, scheduled jobs, custom scripts and a collection of offensive security skills. CloudSEK identified the agent as Strix. Its configuration routed model requests through OpenRouter, with DeepSeek V4.1 Flash handling most of the recorded activity. Gemini 2.5 Flash was configured for delegated sub-agent tasks. The agent had access to several operational tools: - Terminal execution. - Browser and computer interaction. - Scheduled task execution. - Parallel task delegation. - Telegram messaging. - Persistent memory and task tracking. The configuration also included potentially dangerous permissions. For example, certain shell operations and SQL DROP commands were allowed without additional confirmation. The tool-loop hard stop was disabled, allowing the agent to continue after repeated failures. Unsafe JavaScript evaluation in browser sessions was also enabled. The agent's persona instructions encouraged it to operate as an offensive security specialist, minimise safety disclaimers and favour low-noise techniques. ![info-3](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1794472509925-info3--161-.webp) ### How the AI Agent Supported the Operator The conversation records revealed several ways the operator used the system. The operator instructed the agent to: - Analyse reconnaissance output produced by the automated dispatcher. - Review test shell connections. - Investigate whether other parties had accessed the exposed server. - Manage and remove previously acquired offensive security skills. - Continue operational tasks using information stored in persistent memory. The agent also received handoff documentation describing access to the Tajik government document platform. This allowed it to retain operational context and assist with subsequent collection activities. Importantly, the investigation does not establish that the AI agent independently selected government targets or initiated the intrusions. The records instead show a human operator directing an AI-assisted workflow with extensive access to operational tools. ### AI-Assisted Evidence Cleanup CloudSEK also found evidence that the agent had been used to remove operational traces. Its activity included deleting conversation records, rebuilding search indexes, cleaning database records, removing shell histories, clearing cron entries and modifying login records. The cleanup operations took place in early October 2026. However, these efforts did not remove the exposed directory that ultimately revealed the operation. This finding illustrates an important operational risk: AI agents with broad system permissions can execute destructive administrative tasks when their permissions and approval mechanisms are not sufficiently restricted. ## Kazakhstan Nexus and Attribution Findings CloudSEK assessed with high confidence that the exposed server belonged to a Silent Lynx operator. The assessment was based on multiple technical indicators, including malware code markers, development paths, reverse-shell characteristics and delivery techniques consistent with previously documented Silent Lynx activity. The investigation also identified evidence supporting a possible Kazakhstan connection. Although most server logins passed through Tor exit nodes and commercial VPN infrastructure, the records contained multiple connections from residential Kazakh internet providers. The operator's own testing activity also connected to the server through a Kazakhtelecom residential connection before switching to Proton VPN. CloudSEK found that the operator's recorded conversations were conducted in Russian. These findings support previous vendor assessments identifying a possible Kazakhstan nexus. However, network infrastructure and language indicators do not independently establish the operator's identity or nationality. CloudSEK's geographic assessment should therefore be understood as an intelligence assessment rather than confirmed personal attribution. ## Telegram's Role in the Operation Telegram served as an important coordination channel. The operator used a Telegram bot connected to the Hermes agent to issue instructions remotely. The automated document collectors also sent progress updates through Telegram. CloudSEK found that the agent's records linked inbound commands to a single Telegram account, with the display name Marat. This is significant because it demonstrates how a messaging platform can become part of a wider AI-assisted operational environment. The malware implants themselves used TLS-based callback mechanisms, while Telegram provided a separate channel for interacting with the AI agent and monitoring collection activities. ## Operational Security Failures Exposed the Operation Despite the sophistication of its tooling, the Silent Lynx operator made several operational security mistakes. The most important failures included: **1. Centralised storage of sensitive information** Malware source code, victim credentials, stolen documents, network tunnelling information and AI conversation histories were stored on the same server. **2. Reuse of infrastructure** The same server address appeared across malware strings, phishing command lines, PowerShell communications, reconnaissance activities and scheduled callbacks. **3. Exposed development information** Malware binaries retained development paths and project identifiers associated with Laplas and Chvost implants. **4. Poor secrets management** Victim credentials, private keys, API information and access details were stored alongside operational tools. **5. Inconsistent evidence cleanup** Although the operator used the AI agent to delete histories and other records, additional logging processes continued exporting conversations and archiving collected files. **6. Testing activity revealed infrastructure connections** The operator's own test systems reused identifiable host information across different network connections, helping researchers connect activity that was otherwise routed through privacy services. These mistakes allowed CloudSEK to reconstruct significant parts of the operation from information that should not have been publicly accessible. ![info-4](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1794472546501-info4--8-.webp) ## Indicators of Compromise (IOCs) CloudSEK published technical indicators that security teams can use for investigation and threat hunting. Type Indicator Description IPv4 89[.]125[.]213[.]123 Operator infrastructure URL hxxp://89[.]125[.]213[.]123:8080/P.ps1 PowerShell payload URL hxxp://89[.]125[.]213[.]123:8080/decoy.pdf Decoy document Domain foreverpeace[.]ru Associated infrastructure IPv4 134[.]17[.]14[.]24 Smart Forestry cron callback, ownership unconfirmed **Selected SHA-256 hashes:** File SHA-256 srvdriv.exe (Tajikistan build) af2822560bc0e44b600eefbe0170d3526e035704b533a806f9d455da6ac30d2c srvdriv.exe (railway build) f26756cabe92842e25d82bee5f65ef9c22c80aec895386b9840b471b42e01463 netsrvc.exe 5d8a2df90fa3cf913e2b5764e586ee3405a101294909df091060bd3e3d1a44c1 CoreMessaging.exe 19b68e504e1512dc19feef6b1da0f67f876191a937fdcd64ffb965c12d1f0bf0 ssdpdrv.exe 27dc5628adc666da702c9a33e8f96bb87087ac68a7064f99b91fb76b932aef6e Security teams should validate these indicators against the original CloudSEK report before operational use. Infrastructure indicators can change, and an IOC match should be investigated alongside other evidence. ## What Security Teams Can Learn from the Silent Lynx Investigation The investigation offers several practical lessons for organisations operating government, transport and critical infrastructure systems. **Strengthen identity and access management** A standard employee account should not provide access to sensitive government correspondence beyond its authorised scope. Organisations should enforce least-privilege access, strong authentication and regular permission reviews. **Improve application-level authorisation** Authentication alone is not sufficient. Applications must independently verify whether users are authorised to access individual documents, attachments and sensitive records. **Secure legacy infrastructure** The railway intrusion highlights the risks associated with outdated systems, weak SMB configurations, exposed administrative credentials and unpatched servers. Organisations should prioritise asset discovery, vulnerability management, secure configuration and removal of unnecessary legacy services. **Monitor PowerShell and suspicious execution chains** Security teams should investigate unusual shortcut execution, hidden PowerShell processes, remote script downloads and unexpected outbound TCP connections. **Restrict AI agent permissions** AI agents connected to terminals, browsers, databases and messaging platforms should operate with narrowly scoped permissions. Destructive commands should require explicit approval. Organisations should also maintain independent audit logs, restrict access to secrets and monitor agent activity. **Protect sensitive operational information** Malware repositories, credentials, private keys, victim information and investigation records should not be stored together in publicly accessible or insufficiently protected environments. ## Why This Investigation Matters The Silent Lynx investigation highlights a broader shift in how cyber espionage operations are organised. Traditional intrusion techniques, including phishing, reverse shells, credential theft and network reconnaissance, remain central to the operation. However, AI agents are now being incorporated into the surrounding workflow to help coordinate tasks, interpret results and manage operational information. For government agencies and critical infrastructure operators, this creates additional security considerations. The risks are not limited to the initial compromise of an endpoint. A single employee account can potentially expose information across interconnected systems when authorisation controls are weak. Similarly, AI agents with unrestricted access can introduce new operational risks when they are allowed to execute commands, manage credentials and modify system records without appropriate human oversight. The investigation also demonstrates the value of external attack surface monitoring. A poorly secured server exposed information that provided researchers with insight into malware development, victim targeting, infrastructure and operational procedures. **The central lesson is that sophisticated cyber operations can still be exposed through basic security failures.** Strong identity controls, properly segmented infrastructure, secure data storage and carefully governed AI tools remain essential components of an effective cybersecurity strategy. ## Original source https://www.cloudsek.com/blog/off-the-rails-inside-an-exposed-silent-lynx-server? ## Tags `#SilentLynx` · `#CyberEspionage` · `#AIinCybersecurity` · `#DeepSeek` · `#APT` · `#ThreatIntelligence` · `#CriticalInfrastructure` · `#GovernmentCyberSecurity` --- ## About this content This Markdown news article is the citation-grade twin of [Exposed Silent Lynx Server Reveals AI-Assisted Cyber Espionage Against Central Asian Governments](https://www.xcademia.com/news/exposed-silent-lynx-server-reveals-ai-assisted-cyber-espionage-against-central-asian-governments). It is published by **Xcademia** (UK Companies House 12322710) and is available for AI search engines and large language models to index, summarise, and cite. When citing or quoting, please attribute *Xcademia* and link back to the source URL above. - Source: https://www.xcademia.com/news/exposed-silent-lynx-server-reveals-ai-assisted-cyber-espionage-against-central-asian-governments - Publisher: Xcademia, https://www.xcademia.com - Catalogue index: https://www.xcademia.com/llms-full.txt