AI Agent Used in Cyber-Espionage of Tajik Government, Report Says

NewsTree24 | November 12, 2026 — A cybersecurity investigation has documented an espionage operation in which an autonomous AI agent was used to assist reconnaissance, data collection and operational record-keeping during intrusions targeting Central Asian organisations. CloudSEK published its findings on November 11; Reuters reported the investigation the same day.

What the investigation found

CloudSEK said researchers discovered an exposed server containing tools, stolen information and records of an operator’s AI-agent conversations. The company attributed the activity to the Silent Lynx espionage cluster. Reuters reported that CloudSEK found more than 2,000 national-security-related documents collected from Tajikistan’s electronic document-management system.

According to CloudSEK, the operator used the open-source Hermes agent framework, with DeepSeek models handling much of the logged activity. The agent was assigned work such as analysing reconnaissance results, organising collected files and maintaining operational notes. CloudSEK’s report also describes intrusions involving Tajik railway systems and Uzbekistan’s Smart Forestry platform.

How confident is the attribution?

CloudSEK linked the activity to Silent Lynx based on similarities in malware tooling and other technical evidence. It assessed a Kazakhstan connection from factors including Russian-language communications and logins associated with a residential Kazakh internet connection. The company described the group attribution as high confidence and the geographic nexus at a lower confidence level. That is a researcher assessment, not a public confirmation by the governments involved.

Reuters said Tajik and Kazakh officials did not immediately respond to requests for comment. The document count and the description of the AI agent’s role should therefore be attributed to CloudSEK’s investigation rather than treated as independently confirmed government findings.

Why this matters beyond the targeted region

The case is significant because it illustrates how agent frameworks can coordinate multiple routine tasks in an intrusion. The AI system did not remove the need for an operator or make the campaign autonomous in every respect. But it appears to have helped the operator organise work and execute parts of a longer process with less manual interaction.

For enterprises, the same broad capabilities that make agents useful in IT operations—tool access, persistent context, task delegation and the ability to interact with systems—also create risks when permissions and oversight are weak. Security teams should treat agent credentials and tool connectors as privileged access, not as harmless productivity add-ons.

Practical defensive steps for organisations

  • Maintain an inventory of AI agents, their owners, models, connectors and permissions.
  • Give agents narrowly scoped credentials and rotate secrets rather than embedding long-lived access tokens in workflows.
  • Require approval for sensitive actions such as bulk exports, privilege changes and access to regulated records.
  • Log agent instructions, tool calls and data transfers in a way that supports incident investigation.
  • Segment critical systems and validate authorisation at the document or record level, not only at the application login.
  • Monitor unusual outbound transfers and unexpected agent activity, including actions performed through approved automation tools.

What remains uncertain

The public evidence comes primarily from CloudSEK’s analysis of an exposed server and related records. The affected governments have not publicly confirmed the full scope in the reporting reviewed for this article. The case should not be generalised into a claim that AI agents routinely conduct successful cyber-espionage operations; it is a documented investigation that security teams should study carefully.

For related coverage, read NewsTree24’s report on Japan’s AI-backed cybersecurity response and enterprise controls for stopping rogue AI agents.

Sources

Illustrative featured image: Boitumelo via Unsplash.

Leave a Comment