Skip to content

Back to the catalog

UAT-10147: Chinese-speaking adversary integrates agentic AI into post-compromise operations
blog.talosintelligence.comwriteup

UAT-10147: Chinese-speaking adversary integrates agentic AI into post-compromise operations

Cisco Talos details UAT-10147, a Chinese-speaking cybercrime group exploiting one-day vulnerabilities in Windows/Linux web servers to deploy BadIIS, QuasarRAT, NoodleRAT, and SPECTRE implants, while also integrating agentic AI tools (PentestGPT, DeepAudit, and AI-orchestrated ysoserial workflows) to generate exploit playbooks, automate ASP.NET ViewState deserialization RCE, validate exploit success, and run reconnaissance/exfiltration end to end. Talos recovered AI-generated Python scripts and a findings log from the actor's C2 open directory showing semi-autonomous exploit refinement, OOB callback verification via webhook.site, and privilege escalation via SeImpersonatePrivilege/Potato exploits — evidence of a shift from AI-assisted scripting toward agentic offensive orchestration.

TL;DR Talos exposes UAT-10147, a Chinese-speaking actor using PentestGPT/DeepAudit-driven agentic AI to automate ViewState deserialization exploitation, BadIIS deployment, and post-compromise recon at scale.
Document

What it is

Cisco Talos' technical breakdown of UAT-10147's Windows infection chain, with the specific indicators, file names, and defense-evasion commands the summary omits. Discovered in early 2026 via an OPSEC failure: a compromised host beaconing to a download server at 139.180.197[.]150, which exposed an open directory revealing the actor's tooling and a target list of ~170,000 URLs.

How it works

After gaining RCE (or dropping a web shell for manual setup), the actor runs a main batch script named back.txt or back.bat. It uses certutil to pull down EfsPotato (renamed prcc1.rar), a secondary script (bai.bat), and QuasarRAT disguised as svchosts.exe. EfsPotato provides privilege escalation; the script then adds IIS directories to Windows Defender's exclusion list (blinding AV to where malicious IIS modules land), and deletes staging files to hinder forensics. bai.bat establishes persistence via a scheduled task named "Google Chrome Start" that runs the payload with highest privileges at every user logon. Other implants observed in related campaigns: Gh0stCringe and SPECTRE.

Preconditions / Affected

Internet-exposed Windows and Linux web servers with publicly disclosed, unpatched vulnerabilities. Confirmed victims sit in government, universities, media, technology, and gaming; observed compromised servers in Brazil, Bolivia, China, Canada, and Vietnam. The ~170,000-URL target list was split into 17 files of ~10,000 (the actor uses "w" for 萬 = 10,000), indicating broad opportunistic scanning rather than narrow targeting.

What to do

Hunt for the Defender exclusion additions targeting IIS paths — a strong signal, since legitimate admins rarely exclude these:

powershell Add-MpPreference -ExclusionPath C:\Windows\SysWOW64\inetsrv
powershell Add-MpPreference -ExclusionPath C:\Windows\System32\inetsrv

Detections to build:

  • Scheduled task named "Google Chrome Start" running with highest privileges at logon.
  • Dropped/staged file names: back.txt, back.bat, bai.bat, prcc1.rar, svchosts.exe.
  • Network connections to 139.180.197[.]150.
  • certutil used to download the above artifacts.
  • QuasarRAT masquerading as svchosts.exe; monitor System32\inetsrv and SysWOW64\inetsrv for unexpected IIS modules (BadIIS).

Caveats

Talos assesses the agentic-AI orchestration claim (iterative exploit refinement, adaptive troubleshooting, automated post-exploitation) with moderate-to-high confidence, not certainty. The BadIIS deployment steps are described as what the actor "would likely perform," and Gh0stCringe/SPECTRE were seen in "similar campaigns" rather than confirmed in every intrusion. SPECTRE details are in a separate companion post.

Organisations named here

  • Baron Samedit
  • Dirty Pipe
  • SeImpersonatePrivilege
  • TypeConfuseDelegate

CVEs named here

  • CVE-2022-0995
  • CVE-2021-3156
  • CVE-2015-5287
  • CVE-2015-3246
  • CVE-2010-3904
  • CVE-2022-0847
  • CVE-2022-27925
  • CVE-2021-23758
  • CVE-2021-29441
  • CVE-2021-29442
  • CVE-2019-18935

Technologies named here

  • Metasploit Framework
  • Windows web servers
  • Linux web servers
  • ysoserial
  • PentestGPT
  • Zimbra Collaboration Suite
  • AjaxPro
  • DeepAudit
  • EfsPotato
  • Nacos framework
  • certutil
  • Telerik UI for ASP.NET AJAX
  • badsecrets
  • Linux kernel
  • Unix sudo utility
  • appcmd
  • webhook.site
  • ABRT
  • ClamAV
  • libuser
  • IIS
  • Snort
  • .NET 4.8
  • Metasploit
  • .NET

Techniques named here

  • Publicly disclosed vulnerability exploitation
  • AI-driven exploit automation and validation
  • Windows Defender exclusion abuse
  • Scheduled task persistence
  • Rogue local user account creation
  • Web shell deployment
  • ASP.NET ViewState deserialization RCE
  • Out-of-band exploit confirmation
  • Token impersonation privilege escalation
  • Exfiltration blended with legitimate SaaS traffic
  • Two-step web shell deployment via relay upload handler
  • SEO fraud via BadIIS
  • Target list segmentation for scanning efficiency

Threat actors named here

  • UAT-10147
  • BadIIS
  • QuasarRAT
  • Gh0stCringe
  • SPECTRE
  • NoodleRAT
  • Meterpreter