Threat Actor Used Gemini CLI to Build and Operate a Botnet in Minutes

· Original article ↗

Summary

Trend Micro analyzed more than 200 session logs showing a Russian-speaking actor using Gemini CLI to migrate and operate a botnet targeting eight dental-clinic computers, alongside credential attacks and fraud planning.

Key points

  • The actor used Gemini CLI to migrate a live C2 server and restore bot connections in minutes, with the AI handling coding, deployment, configuration, and debugging.
  • The botnet controlled eight computers at a dental clinic and accessed its OpenDental database; victims beaconed to the C2 over HTTPS and received PowerShell commands.
  • The C2 operation was documented in three plain-text files totaling roughly 5 KB, making it easy to redeploy or adapt.
  • The actor also used AI to generate password variants for WordPress brute-force attacks, process stolen credentials, and plan cryptocurrency fraud; some WordPress compromises succeeded.
  • The malware used five-second polling, PowerShell payloads, and persistence through WMI subscriptions or scheduled tasks, depending on the actor's privileges.
  • Trend Micro recommends behavior-based detection, including monitoring recurring outbound polling, PowerShell from unusual locations, and runtime-created WMI subscriptions.
  • The article reports no definitive takedown or remediation status; it warns that rapid rebuilding could reduce the impact of server takedowns.

Article Details

Attack Vectors
  • The actor used Gemini CLI to deploy and operate a command-and-control server and botnet, with victims beaconing over HTTPS and polling for PowerShell commands.
  • A PowerShell stager downloaded agent_final.ps1 from payloads.tralalarkefe[.]com.
  • The botnet used recurring HTTP GET polling every five seconds, Base64-encoded tasking and telemetry, and custom HTTP headers identifying infected machines.
  • Persistence was established through WMI event subscriptions and a startup scheduled task with administrator privileges, or a logon registry value and scheduled task without administrator privileges.
  • The actor used AI-generated password variants in brute-force attempts against WordPress admin panels, with some reported successes.
  • The actor processed stolen credentials and cookies to attempt access to a victim company's VPN and internal admin panel and to compromise an eCommerce site.
Defensive Notes
  • Prioritize behavioral detection over static indicators, including recurring outbound polling, PowerShell running from non-standard locations, and WMI subscriptions created at runtime.
  • Enforce unique passwords, monitor employee credentials against breach databases, and require phishing-resistant multi-factor authentication.
  • Pair server takedowns with network-level blocking and ongoing monitoring for reconnection attempts.

Indicators of compromise

TypeIndicatorContext
DOMAINpayloads[.]tralalarkefe[.]comPayload-delivery domain used to download the botnet's PowerShell agent.

MITRE ATT&CK

People

Threat Actors

Vendors

Products

Tools

Countries

Industries

Related Articles