Skip to navigation

Threat reportAPTTL-2026-6164

ARTEX AI Pentesting Agent Weaponized by China-nexus VAPOR PANDA and North Korean Onyx Sleet (Andariel)

highACTIVE

ARTEX, an open-source Go-based autonomous multi-agent pentest/C2 framework, is being abused by multiple threat actors. CrowdStrike reports VAPOR PANDA (China-nexus) has used it since 2026-10-17 for parallel intrusions against 200+ organizations in at least 14 countries (post-exploitation confirmed at nine), and Microsoft reports Onyx Sleet (Andariel) and Pearl Sleet have used it since late October 2026. First disclosed .

Key facts

IdentifiersTL-2026-6164
Also known asARTEX, ARTEX PRO, ScopeWeaver
SeverityHIGH
First disclosed
Affected productsMultiple Internet-facing enterprise applications, cloud identity/IaaS (AWS, Azure), Active Directory, SAML-based SSO, VPN
Weaknesses (CWE)CWE-611, CWE-347, CWE-918, CWE-798
AttributionVAPOR PANDA, Silent Chollima (medium confidence)
Nation-state nexusChina, North Korea
MalwareBloodAlchemy
ToolsARTEX, Chisel, SoftEther, Tor - S0183
MITRE ATT&CK techniquesT1003.006 OS Credential Dumping: DCSync, T1036.005 Masquerading: Match Legitimate Resource Name or Location, T1078 Valid Accounts, T1090.003 Proxy: Multi-hop Proxy, T1110.003 Brute Force: Password Spraying, T1114.002 Email Collection: Remote Email Collection, T1133 External Remote Services, T1134.001 Access Token Manipulation: Token Impersonation/Theft, T1190 Exploit Public-Facing Application, T1505.003 Server Software Component: Web Shell, T1518.001 Software Discovery: Security Software Discovery, T1552.005 Unsecured Credentials: Cloud Instance Metadata API and 8 more in the attack flow
MotivationESPIONAGE
Target sectorsgovernment administration, think-tank, pharmacy, semi-conductors, defense, nuclear, finance, telecoms, energy, enterprise-software, industrial-automation, chemicals
Target regionsNorth America, Europe, Latin America, Middle East, south korea, Asia, Oceania
Detection rules9 (1 shown free)
Sources8 cited sources

Detection coverage

As of 2026-11-16, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-6164 across Splunk SPL, Microsoft KQL and Sigma, covering 21 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: BLOODALCHEMY GameHook svchost.exe Side-Loading of graphics-hook-filter64.dll (ARTEX / VAPOR PANDA)

Detects the BLOODALCHEMY loader chain delivered by VAPOR PANDA after ARTEX-driven access: a legitimate GameHook binary renamed svchost.exe that side-loads graphics-hook-filter64.dll and decrypts graphics-hook_config.bin. Fires on the loader DLL load, the payload file drop, the known renamed-binary/loader hashes, or any svchost.exe running or loading modules from outside the Windows system directories. Behaviour-first, so it survives hash rotation.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1574.001, T1036.005
  • Data sources: sysmon, microsoft_defender_for_endpoint
  • Severity: critical
  • Confidence: high

Splunk SPL

index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" ((EventCode=7 (ImageLoaded="*\\graphics-hook-filter64.dll" OR Hashes="*3FB5A9FB3E2CB1D6208966BD4E55D10282D19C7250D827EFA560463D60A12EE4*")) OR (EventCode=11 TargetFilename="*\\graphics-hook_config.bin") OR (EventCode IN (1,7) (Hashes="*49FE70419FD19B81CA8D730A6EF7C1EC71B681D60B390761070D42686C7D2068*" OR (Image="*\\svchost.exe" NOT (Image="C:\\Windows\\System32\\svchost.exe" OR Image="C:\\Windows\\SysWOW64\\svchost.exe" OR Image="C:\\Windows\\WinSxS\\*")))))
| eval indicator=case(like(lower(ImageLoaded),"%graphics-hook-filter64.dll"),"bloodalchemy_loader_dll_load", like(lower(TargetFilename),"%graphics-hook_config.bin"),"bloodalchemy_payload_file", like(lower(Hashes),"%49fe7041%"),"gamehook_renamed_svchost_hash", true(),"svchost_outside_system_paths")
| stats count min(_time) as first_seen max(_time) as last_seen values(indicator) as indicators values(ImageLoaded) as loaded_module values(Image) as process_image values(ParentImage) as parent_image values(CommandLine) as command_line by host, User
| convert ctime(first_seen) ctime(last_seen)

Microsoft KQL

let LoaderHashes = dynamic(['3fb5a9fb3e2cb1d6208966bd4e55d10282d19c7250d827efa560463d60a12ee4','49fe70419fd19b81ca8d730a6ef7c1ec71b681d60b390761070d42686c7d2068','364251838edfcf5ef1107fa692da3c5bd9a540482449f645992830134bf807ab']);
union
(DeviceImageLoadEvents
 | where Timestamp > ago(24h)
 | where FileName =~ 'graphics-hook-filter64.dll' or SHA256 in~ (LoaderHashes)
 | project Timestamp, DeviceName, AccountName=InitiatingProcessAccountName, Signal='bloodalchemy_loader_dll_load', FileName, FolderPath, SHA256, InitiatingProcessFileName, InitiatingProcessFolderPath, CommandLine=InitiatingProcessCommandLine),
(DeviceFileEvents
 | where Timestamp > ago(24h)
 | where ActionType == 'FileCreated'
 | where FileName =~ 'graphics-hook_config.bin' or SHA256 in~ (LoaderHashes)
 | project Timestamp, DeviceName, AccountName=InitiatingProcessAccountName, Signal='bloodalchemy_payload_file', FileName, FolderPath, SHA256, InitiatingProcessFileName, InitiatingProcessFolderPath, CommandLine=InitiatingProcessCommandLine),
(DeviceProcessEvents
 | where Timestamp > ago(24h)
 | where SHA256 in~ (LoaderHashes) or (FileName =~ 'svchost.exe' and not(tolower(FolderPath) startswith @'c:\windows\system32\' or tolower(FolderPath) startswith @'c:\windows\syswow64\' or tolower(FolderPath) startswith @'c:\windows\winsxs\'))
 | project Timestamp, DeviceName, AccountName, Signal='svchost_masquerade_or_gamehook_hash', FileName, FolderPath, SHA256, InitiatingProcessFileName, InitiatingProcessFolderPath, CommandLine=ProcessCommandLine)
| order by Timestamp desc

Sigma

title: BLOODALCHEMY GameHook svchost.exe Side-Loading graphics-hook-filter64.dll
id: 7c1f3a52-9d84-4b6e-a0c5-3e2d81f4b7a1
status: experimental
description: Detects the BLOODALCHEMY loader chain used by VAPOR PANDA in ARTEX-driven intrusions, a legitimate GameHook binary renamed svchost.exe side-loading graphics-hook-filter64.dll, or any svchost.exe loading modules from outside the Windows system directories.
references:
    - https://www.crowdstrike.com/en-us/blog/vapor-panda-uses-artex-to-scale-global-campaign/
    - https://www.picussecurity.com/resource/blog/artex-in-the-wild-how-threat-actors-weaponize-an-ai-pentesting-agent
author: Threadlinqs Team
date: 2026/11/16
tags:
    - attack.defense-evasion
    - attack.t1574.001
    - attack.t1036.005
logsource:
    category: image_load
    product: windows
detection:
    selection_dll:
        ImageLoaded|endswith: '\graphics-hook-filter64.dll'
    selection_hash:
        Hashes|contains:
            - '3FB5A9FB3E2CB1D6208966BD4E55D10282D19C7250D827EFA560463D60A12EE4'
            - '49FE70419FD19B81CA8D730A6EF7C1EC71B681D60B390761070D42686C7D2068'
    selection_svchost:
        Image|endswith: '\svchost.exe'
    filter_legit_paths:
        Image|startswith:
            - 'C:\Windows\System32\'
            - 'C:\Windows\SysWOW64\'
            - 'C:\Windows\WinSxS\'
    condition: selection_dll or selection_hash or (selection_svchost and not filter_legit_paths)
falsepositives:
    - Legitimate GameHook or game-overlay software installed alongside a renamed binary in a lab
    - Software packaging tools that stage a copy of svchost.exe outside system directories
level: critical

Known false positives:

  • Legitimate game-overlay or GameHook software on a developer or lab workstation
  • Software packaging or sandbox tooling that stages a copy of svchost.exe outside the Windows system directories

Tuning notes: Requires Sysmon EventCode 1, 7 and 11 (image-load logging enabled for svchost.exe). Adjust the index name to your Sysmon index. If WinSxS servicing causes noise, scope the NOT clause to hash-verified Microsoft binaries. The loader DLL name and hashes are stable IOCs, the svchost path check is the behavioural backstop.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • Web Server Process Querying Cloud Instance Metadata Service (SAML/XXE to IMDS Credential Theft) · unlock with TL Creds
  • ARTEX Framework Hunting Markers (artx5v, [indicator], ArtexProbe, art18-) · unlock with TL Creds
  • Tunneling and Proxy Tooling Execution (Chisel, SoftEther VPN, Tor) by VAPOR PANDA · unlock with TL Creds
  • JSP Web Shell Written by MySQL General Log Poisoning ([indicator]) · unlock with TL Creds
  • LAPS Password Retrieval and RBCD Delegation Setup in Active Directory (ARTEX Post-Exploitation) · unlock with TL Creds
  • curl or wget Beaconing and Exfiltration to [indicator] and OAST Services (ARTEX C2) · unlock with TL Creds
  • DCSync Replication Rights and Kerberoasting Burst (ARTEX Active Directory Post-Exploitation) · unlock with TL Creds
  • BLOODALCHEMY C2 Infrastructure Communication ([indicator], [indicator], ddduuuppp.com) · unlock with TL Creds

Indicators of compromise

As of 2026-11-16, Threadlinqs Intelligence tracks 21 indicators of compromise for TL-2026-6164.

Sample indicators (3 of 21)

  • IP address: 45.153.126[.]223
  • Domain: nginxmysql[.]com
  • File hash: 49fe70419fd19b81ca8d730a6ef7c1ec71b681d60b390761070d42686c7d2068

Defanged, TLP:CLEAR. Full set of 21 indicators: unlock with TL Creds.

21 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

ARTEX is an open-source, LLM-driven autonomous multi-agent penetration-testing and offensive C2 framework written in Go with an embedded Next.js web console. A main agent coordinates planner and worker agents (six specialized sub-agents) around a shared graph of assets, findings and goals, and workers execute standard offensive tooling (nmap, curl, wget, Impacket, Python HTTP clients, headless Chromium via Playwright). Any LLM backend can be used. It was published in July 2026 by a Chinese developer using the GitHub handle Autumn-27, won Baidu's Agent+ red-team challenge in September 2026, and was closed-sourced on 2026-10-08 after more than 12,000 forks; community forks include ScopeWeaver and the commercial ARTEX PRO. The underlying techniques are not new; the framework compresses the time between attack steps and lets one operator run many intrusions in parallel.

CrowdStrike attributes a campaign beginning 2026-10-17 to VAPOR PANDA, a China-nexus espionage actor active since 2021 against government, think tank, energy and extractive targets. A single ARTEX controller drove intrusions against 200+ organizations in at least 14 countries within an 11-day window, using DeepSeek as the primary LLM plus a second commercial LLM. Post-exploitation (access, BLOODALCHEMY deployment, credential spraying, scanning, reconnaissance) was confirmed at nine organizations, with spraying/scanning only at 12+ more. Victims include foreign and interior ministries in Latin America, the Middle East and Europe (including a Greek government entity), 19 US organizations in pharmaceutical, semiconductor, defense, nuclear and financial sectors, and telecom providers in the US, Belgium and the Philippines. A human operator issued Chinese-language instructions timestamped in China Standard Time; planning agents in one intrusion consumed roughly 1.4 billion input tokens. Reported tradecraft includes unsigned SAML assertion plus XXE chaining to query the cloud instance metadata service, cloud role abuse and escalation to enterprise admin, JSP/PHP web shells, abuse of an unauthenticated orchestration service for root in a container environment, LAPS password reads, resource-based constrained delegation, forged mail-admin tokens from unrotated keys in a 2022 directory export, reuse of VPN configs and keys, an EDR coverage check for Windows hosts without Falcon, and BLOODALCHEMY delivered by a renamed GameHook binary (svchost.exe) side-loading a DLL. Proxying used Tor, SOCKS, Chisel and SoftEther. CrowdStrike states this activity is distinct from the ARTEX activity Microsoft tracks as Storm-0940 (Google UNC7228). Picus's threat library maps VAPOR PANDA to Salt Typhoon; CrowdStrike's report does not make that statement.

Microsoft reports that North Korean actors Onyx Sleet (Andariel) and Pearl Sleet have used ARTEX since late October 2026, and that Onyx Sleet produced 100+ target-specific, likely AI-generated PDF lures in August-September 2026 for high-volume phishing of South Korean users. Picus notes DPRK AI-enabled operations remain human-directed. Google GTIG/Mandiant documents separate ARTEX use (UNC7228/Storm-0940, financially motivated extortion groups and other suspected PRC clusters) from late August 2026: Azure/AWS permission mapping, Kudu abuse, webhook.site-based C2 and exfiltration (491 rotating URLs in one incident), Kerberoasting, DCSync, MySQL general-log web shell writing, and ephemeral EC2 instances for EBS harvesting. Single-vendor-derived items (Picus) are corroborated by CrowdStrike, Microsoft and Google reporting; no CVE is assigned and severity is analyst-assigned.

MITRE ATT&CK techniques by tactic:

Credential Access

T1003.006 OS Credential Dumping: DCSync; T1110.003 Brute Force: Password Spraying; T1552.005 Unsecured Credentials: Cloud Instance Metadata API; T1555 Credentials from Password Stores; T1558.003 Steal or Forge Kerberos Tickets: Kerberoasting; T1606.002 Forge Web Credentials: SAML Tokens

Defense Evasion

T1036.005 Masquerading: Match Legitimate Resource Name or Location; T1574.001 DLL

Initial Access

T1078 Valid Accounts; T1190 Exploit Public-Facing Application; T1566.001 Phishing: Spearphishing Attachment

Command and Control

T1090.003 Proxy: Multi-hop Proxy; T1572 Protocol Tunneling

Collection

T1114.002 Email Collection: Remote Email Collection

Persistence

T1133 External Remote Services; T1505.003 Server Software Component: Web Shell

Privilege Escalation

T1134.001 Access Token Manipulation: Token Impersonation/Theft

Discovery

T1518.001 Software Discovery: Security Software Discovery

Resource Development

T1588.007 Obtain Capabilities: Artificial Intelligence

Execution

T1651 Cloud Administration Command

Remediation

Patches

  • Patch internet-facing appliances and long-lived Linux servers, including firmware

Immediate actions

  • Hunt for ARTEX markers: command lines containing artx or artex, artx-prefixed cloud roles, art18- comments in cloud administration commands, directory C:\ProgramData\artx5v\, artex-probe-* files and accounts containing ArtexProbe
  • Block the BLOODALCHEMY C2 domains and IPs listed in the IOCs and search proxy/DNS logs for them
  • Rotate signing keys, preauthentication keys and service credentials after any suspected exposure, including keys in historical backups and directory exports
  • Alert on webhook.site, OAST and tunneling egress and on curl-driven script fetching from database or server hosts
  • Verify EDR sensor coverage on all servers including those not internet reachable

Workarounds

  • Isolate web-facing systems behind a VPN, WAF, reverse proxy or DMZ
  • Add rate and recipient limits to high-impact APIs (email, notifications, credential retrieval, user creation)

Longer-term hardening

  • Enforce SAML signature validation and require IMDSv2
  • Limit cloud role command permissions to the minimum necessary instances and move to short-lived credentials (IAM roles, Managed Identities, Workload Identity Federation)
  • Harden Active Directory: Credential Guard, Protected Users, SMB/LDAP signing, channel binding, LAPS access review, RBCD review, AD CS template hardening
  • Restrict admin APIs, consoles and CI/CD (including Kudu/SCM) to trusted networks; restrict database runtime privileges to block OS command execution and arbitrary file writes
  • Baseline identity origin (ASN, provider, region) and alert on control-plane bursts, storage enumeration and bulk downloads
  • Deploy phishing defenses (Safe Links, Safe Attachments, ZAP) and verify hire identities against DPRK IT-worker risk

Affected and fixed versions

  • Multiple — Internet-facing enterprise applications, cloud identity/IaaS (AWS, Azure), Active Directory, SAML-based SSO, VPN
    Vulnerable versions: Environments with unrotated signing keys, weak SAML validation, IMDSv1, exposed admin/orchestration services

Timeline

  • ARTEX open-source Go-based multi-agent pentest/C2 framework published by developer Autumn-27 (July 2026; exact day not stated).
  • Earliest observed ARTEX abuse (late August 2026, per Google GTIG, UNC7228/Storm-0940); Onyx Sleet begins generating 100+ AI-made PDF lures in August-September (day approximate).
  • ARTEX wins Baidu's Agent+ red-team challenge (September 2026; day approximate).
  • Original ARTEX repository closed-sourced after 12,000+ forks; Reuters links the move to a Korean bank hack.
  • Reuters reports the developer closed ARTEX after a Korean bank hack.
  • VAPOR PANDA begins parallel ARTEX-driven intrusions using a single controller; 200+ organizations in 14+ countries within 11 days.
  • Microsoft publishes Storm-0940 reporting covering separate ARTEX use (cited by Google GTIG).
  • Onyx Sleet (Andariel) and Pearl Sleet begin using ARTEX (late October 2026 per Microsoft; day approximate).
  • CrowdStrike publishes VAPOR PANDA/ARTEX report and Microsoft publishes North Korean AI-adoption report.
  • Google GTIG/Mandiant publishes 'ARTEX and friends' with TTPs and defensive layers.
  • Picus Security publishes 'ARTEX in the Wild' consolidating VAPOR PANDA and Onyx Sleet activity.

Sources cited

More from VAPOR PANDA

Drafted by the Threadlinqs research pipeline from 8 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats