Skip to content
Hackers use autonomous AI agents to steal Tajik government documents

Hackers use autonomous AI agents to steal Tajik government documents

Dig.Watch • November 12, 2026

CloudSEK identified Silent Lynx as the group behind the theft of more than 2,000 sensitive Tajik government documents.

In an investigation published on 11 November 2026 by Reuters , Singapore-based cybersecurity company CloudSEK reported that a hacking group identified as Silent Lynx used Hermes , an open-source framework for autonomous AI agents, to obtain more than 2,000 sensitive documents from Tajikistan’s government electronic document management system.

The stolen material reportedly included documents relating to national security and correspondence associated with the Ministry of Defence of Tajikistan. The discovery followed an operational security failure: the attackers inadvertently left a server exposed to the internet, allowing researchers to retrieve instructions, operational records and other material that helped reconstruct the campaign. The incident provides unusually detailed insight into how threat actors are incorporating AI agents into cyber-espionage operations.

According to CloudSEK, Silent Lynx began using Hermes for reconnaissance around mid-August 2026, before accessing Tajikistan’s electronic document system several weeks later. The agent was reportedly instructed to operate as an offensive security specialist and red teamer. Rather than manually issuing individual commands, the operator delegated tasks to the framework, which could plan activities, execute instructions, maintain operational notes and prepare reports. It also reportedly developed a customised programme to retrieve documents, organise the collected files and divide them into smaller batches to facilitate their extraction.

CloudSEK attributed the activity to Silent Lynx, a group that researchers associate with Kazakhstan . Its assessment reportedly draws on several indicators, including Russian-language communications, working hours corresponding to the Kazakh business day and connections to the exposed infrastructure through a residential internet connection in Kazakhstan. These indicators support a geographical assessment but do not establish the identity of the operators or demonstrate state sponsorship. Reuters reported that neither Kazakh nor Tajik government officials had immediately responded to requests for .

The exposed server contained instructions defining the agent’s role and operational behaviour, providing researchers with evidence of how human operators structured and supervised its activities. CloudSEK’s wider investigation also identified compromises affecting Tajikistan’s railway infrastructure and access to Uzbekistan’s Smart Forestry platform . It found preparations for phishing operations impersonating Turkmen government entities, although these preparations should not be confused with confirmed successful attacks.

The technical significance of this case lies in the degree of operational delegation. Hermes is an open-source framework designed to coordinate AI agents capable of carrying out multi-step tasks through connected tools and applications. In this case, the framework reportedly supported reconnaissance, information collection, software development and the organisation of stolen material. CloudSEK’s technical findings further identify DeepSeek-powered agents in the operation, with additional models reportedly used for delegated tasks. Nevertheless, an important distinction must be maintained: the available evidence does not establish that the AI agent independently discovered the vulnerabilities or obtained initial access to the compromised government system.

The investigation indicates that access to the electronic document platform was achieved using an employee’s existing credentials. The AI’s documented contribution therefore appears to have centred on conducting and organising activities after access had been established, rather than independently breaching the system.

The Silent Lynx operation marks also an important development in the relationship between artificial intelligence and cyber espionage. examples of malicious AI use have frequently focused on phishing content, social engineering, malware development or automated vulnerability discovery. This case illustrates a broader operational application: delegating connected stages of an intrusion and intelligence-collection workflow to an AI agent. Such capabilities could reduce the time and human resources required to conduct reconnaissance, process large volumes of information and maintain prolonged access to compromised environments.

The Silent Lynx operation provides a particularly revealing example: the attackers reportedly used Hermes not merely to generate code or support reconnaissance, but to coordinate activities, retrieve sensitive documents and organise the stolen information. Though, no available evidence establishes firmly that the AI agent itself obtained initial access to the compromised government system and independently hacked a national government.

Extracted Entities

Attack Types (1)

Domains (1)

Industries (1)

Malware (1)

MITRE ATT&CK (1)