A Chinese-speaking cybercrime outfit called UAT-10147 is using artificial-intelligence tools to automate large-scale attacks on Windows and Linux web servers. It deploys a new cross-platform implant named SPECTRE and a Linux kernel rootkit dubbed Specter. The group spikes activity in Brazil, Bolivia, China, Canada and Vietnam, hitting multiple industry sectors worldwide.

Why AI matters to attackers

AI-driven assistants such as PentestGPT and DeepAudit run autonomous vulnerability scans and suggest exploit paths. UAT-10147 feeds them target IP ranges, letting the AI handle reconnaissance, exploit selection and payload generation without human hands-on work. This automation shrinks the time from discovery to compromise to minutes, letting the gang strike many servers at once.

The attack chain in practice

  1. Remote code execution (RCE) – The AI spots known flaws in web or IIS servers and triggers RCE, gaining a foothold.
  2. Malware drop – Payloads such as BadIIS, Quasar RAT and Gh0stCringe are installed. BadIIS runs as a Windows service, indicating coordination among Chinese-speaking groups.
  3. Cross-platform implant – The newly observed SPECTRE implant works on both Windows and Linux, hides in memory, steals credentials and injects code into legitimate processes.
  4. Privilege escalation – On Windows the group runs EfsPotato; on Linux it installs the kernel-level rootkit Specter, which gives attackers control at the kernel level and blinds most security software.
  5. Persistence – Fake scheduled tasks survive reboots while the original malicious files are deleted.
  6. Defender bypass – The attackers add exclusions to Microsoft Defender and erase traces, effectively neutralising built-in endpoint protection.

SPECTRE and Specter: why they’re alarming

SPECTRE’s ability to run on multiple operating systems makes it a “one-size-fits-all” backdoor. It abuses vulnerable drivers such as RTCore64.sys to evade endpoint detection-and-response (EDR) solutions, a technique once seen only in sophisticated espionage campaigns. Specter lives in the Linux kernel, giving attackers unrestricted access and rendering most user-space monitoring tools blind. Researchers suspect AI helped write Specter’s low-level code, a development that could lower the barrier for future kernel-level threats.

Who stands to lose

Enterprises that expose web services to the internet—e-commerce platforms, SaaS providers, cloud-hosted applications—are prime targets. A successful compromise can lead to data theft, ransomware extortion, or the use of the server as a launchpad for further attacks. Because the implants slip past conventional antivirus and EDR, detection costs soar; victims may need to rebuild systems from scratch or hire specialized incident-response teams.

The other side of the AI coin

The same AI assistants that accelerate UAT-10147’s operations are marketed for legitimate penetration testing. Security teams can harness PentestGPT-style tools to discover vulnerabilities before attackers do, provided they apply strict controls and verify outputs. The dual-use nature of these AI models fuels a debate: should regulators curb their availability, or should the industry focus on defensive AI that can spot AI-generated malicious code?

What to watch for

  • Unusual scheduled tasks that reference unknown binaries.
  • New exclusions in Microsoft Defender or other AV products.
  • Loading of rarely used drivers, especially those matching known vulnerable signatures.
  • Sudden spikes in outbound traffic from web servers to unknown IP ranges.
  • Presence of unknown services named after known malware families (e.g., BadIIS).

Takeaway

UAT-10147 shows AI is no longer a niche aid for hobbyist hackers; it is a force multiplier for organized crime, enabling rapid, cross-platform compromises at scale. Defenders must match that speed with AI-enhanced monitoring, stricter change-control on server configurations, and readiness to rebuild compromised infrastructure when stealthy implants like SPECTRE slip through.