Emerging threat

ARTEX in the Wild: How Threat Actors Weaponize an AI Pentesting Agent

Since late August 2026, threat actors have used ARTEX, an open-source framework in which LLM-driven agents carry out a penetration test from reconnaissance to data exfiltration, in real intrusions. Two of them are VAPOR PANDA, a China-nexus espionage group that used ARTEX to run parallel intrusions against more than 200 organizations [1], and Onyx Sleet (Andariel), a North Korean group that added it to operations already accelerated by AI [2].

This post explains what ARTEX is, how each of these two actors has used it in its operations, and how to safely simulate the attack techniques and malware used by these threat actors with Picus to test whether your security controls prevent and detect these threats.

What Is ARTEX?

ARTEX is an autonomous, multi-agent penetration testing and offensive command-and-control (C2) platform. It is written in Go and ships with an embedded Next.js web console.

A Chinese developer (GitHub user Autumn-27) published ARTEX in July 2026, and it won Baidu's "Agent+" red team challenge that September. Threat actors picked it up within weeks. The developer closed the source of the original repository on October 8, 2026 [3]. By then the code had more than 12,000 forks, so the code remains publicly available and keeps spreading through community forks, including localized variants such as ScopeWeaver and a commercial version called ARTEX PRO [4].

The operator supplies a target and an objective. From there the framework works roughly as follows:

  • A main agent coordinates subordinate planner and worker agents, with the attack lifecycle split across six specialized sub-agents.
  • The planner chooses the next step. Workers run ordinary offensive tooling: nmap, curl, wget, Impacket, Python HTTP clients, Norma SDK WebFetch, and headless Chromium through Playwright.
  • Every result is written to a shared graph of assets, findings and goals, and the planner reads that graph to decide what to do next.
  • ARTEX can use any LLM backend. One espionage actor ran it primarily on DeepSeek, with a second commercial LLM as a supplement.

The techniques ARTEX runs, such as credential abuse, password spraying, web shells, Kerberoasting and DCSync, have been in use for years. ARTEX shortens the time between them: the agent reads the output of one action and launches the next in seconds, tries several attack paths in parallel, and keeps going around the clock.

Threat Actors Using ARTEX: VAPOR PANDA and Onyx Sleet

VAPOR PANDA: Parallel Espionage Engagements

VAPOR PANDA is a China-nexus espionage adversary that has pursued long-term access to government and think tank networks since 2021. Beginning October 17, 2026, it used a single ARTEX controller, running mainly on DeepSeek with a second commercial LLM, to run intrusions against more than 200 organizations in at least 14 countries at the same time [1]. This was its first observed use of AI-driven intrusion tooling, and it led to confirmed post-exploitation at nine organizations.

The controller intermittently exposed tooling, target files and ARTEX agent transcripts, which show how the operation ran. The operator ran several ARTEX engagements (called "tasks") at once and stepped in only at decision points. Operator instructions in the transcripts are written in Chinese, labeled "human instruction", and timestamped in China Standard Time. In a single intrusion, the planning agents consumed on the order of 1.4 billion input tokens [1].

Examples from the transcripts:

  • US pharmaceutical company #1: The agent chained an unsigned SAML assertion with an XXE flaw to query the cloud Instance Metadata Service. That returned a role able to issue administration commands to thousands of instances, and the agent escalated from there to an enterprise administrator account.
  • US pharmaceutical company #2: The agent deployed web shells, abused an unauthenticated orchestration service to get root in a production container environment, read LAPS passwords, and used resource-based constrained delegation to impersonate a privileged account.
  • Greek government entity: The operator gave the agent a 2022 directory export from the organization's mail server. The authentication keys in it had likely never been rotated, and the agent used them to forge an administrator token and read mailboxes. It then combined VPN configurations from a user's sent mail with a VPN client key the operator said came from a May 2024 email leak to reach the internal network.
  • Global pharmaceutical company: The operator told the agents in Chinese to "find Windows machines in the domain without Falcon, judging by tasklist and folder listings." The operator was building EDR coverage checks into the agent's workflow so follow-on activity would land on unprotected hosts.

Onyx Sleet (Andariel): ARTEX in North Korean Operations

Onyx Sleet, also known as Andariel, is a North Korean threat actor. Like other North Korean groups, it works toward the regime's goals of generating revenue and collecting strategic intelligence, and its tradecraft rests on social engineering, malware delivery and the abuse of legitimate tools. Since late October 2026, Onyx Sleet has used ARTEX in its operations, alongside another North Korean actor, Pearl Sleet. Its overall AI adoption sits at the "accelerating" stage, meaning it is quickly bringing AI into several parts of its operations [2].

Before adopting ARTEX, the group was already using AI to scale its phishing. In August and September 2026, it produced more than 100 PDF lure files with target-specific details, creating dozens of them within minutes, most likely with an AI platform. These lures fed high-volume phishing campaigns against South Korean users, and the speed at which new lures can be generated makes the campaigns harder to detect and disrupt. Across North Korean threat actors, AI-enabled operations, including those using ARTEX, remain human-directed rather than fully autonomous.

How Picus Simulates Attacks by Threat Actors Using ARTEX?

We strongly suggest simulating attacks by threat actors using ARTEX to test the effectiveness of your security controls against real-life cyber attacks using the Picus Platform. You can also test your defenses against hundreds of other threat groups within minutes with a 14-day free trial of the Picus Platform.

Picus Threat Library includes the following threats for Salt Typhoon (VAPOR PANDA) and Andariel (Onyx Sleet):

Threat ID

Threat Name

Attack Module

38581

Salt Typhoon Threat Group Campaign Malware Download Threat

Network Infiltration

28940

Salt Typhoon Threat Group Campaign Malware Email Threat

E-mail Infiltration

26880

Salt Typhoon Threat Group Campaign

Windows Endpoint

70388

Andariel Threat Group Campaign Malware Download Threat

Network Infiltration

71059

Andariel Threat Group Campaign Malware Email Threat

E-mail Infiltration

38883

Andariel Threat Group Campaign

Windows Endpoint

56407

Andariel Threat Group Campaign Backdoor Malware Download Threat

Network Infiltration

41382

Andariel Threat Group Campaign Backdoor Malware Email Threat

E-mail Infiltration

37270

Andariel Threat Group Campaign RAT Download Threat

Network Infiltration

30592

Andariel Threat Group Campaign RAT Email Threat

E-mail Infiltration

76699

Andariel Threat Group Campaign Malware Downloader Download Threat

Network Infiltration

93568

Andariel Threat Group Campaign Malware Downloader Email Threat

E-mail Infiltration

88521

Andariel Threat Group Campaign Ransomware Download Threat

Network Infiltration

38855

Andariel Threat Group Campaign Ransomware Email Threat

E-mail Infiltration

Start simulating emerging threats today

Get actionable mitigation insights with a 14-day free trial of the Picus Platform.

Start free trial

References

[1] CrowdStrike, "VAPOR PANDA uses AI-driven ARTEX to grow global espionage campaign," CrowdStrike Blog. Accessed: Nov. 16, 2026. [Online]. Available: https://www.crowdstrike.com/en-us/blog/vapor-panda-uses-artex-to-scale-global-campaign/

[2] Microsoft Threat Intelligence, "Tracking AI adoption in North Korean threat actor tradecraft," Microsoft Security Blog, Oct. 29, 2026. Accessed: Nov. 16, 2026. [Online]. Available: https://www.microsoft.com/en-us/security/blog/2026/10/29/tracking-ai-adoption-in-north-korean-threat-actor-tradecraft/

[3] Reuters, "Chinese developer makes ARTEX AI agent closed-source after Korean bank hack," Reuters, Oct. 9, 2026. Accessed: Nov. 16, 2026. [Online]. Available: https://www.reuters.com/world/china/chinese-developer-makes-artex-ai-agent-closed-source-after-korean-bank-hack-2026-10-09/

[4] A. Blum, D. Mainor, R. Nambiar, O. ElAhdan, M. McWhirt, W. Silverstone, B. Dhake, and C. Busse, "ARTEX and friends: Enter agentic offensive frameworks," Google Cloud Blog. Accessed: Nov. 16, 2026. [Online]. Available: https://cloud.google.com/blog/topics/threat-intelligence/artex-and-friends-enter-agentic-offensive-frameworks

Resources

Latest Resources

  • ARTEX in the Wild: How Threat Actors Weaponize an AI Pentesting Agent

    ARTEX in the Wild: How Threat Actors Weaponize an AI Pentesting Agent

  • Node2Shell (CVE-2026-111497): Node.js Brotli RCE Vulnerability Explained

    Node2Shell (CVE-2026-111497): Node.js Brotli RCE Vulnerability Explained

  • Famous Chollima APT Group: Major Activities, TTPs, and Simulation

    Famous Chollima APT Group: Major Activities, TTPs, and Simulation

  • What PHP2Shell Reveals About the Limits of Automated Pentesting

    What PHP2Shell Reveals About the Limits of Automated Pentesting

  • PHP2Shell: Critical PHP RCE Flaws (CVE-2026-109618 & CVE-2026-109619) Explained

  • CVE-2026-88771: Citrix NetScaler Pre-Auth Command Injection Explained

    CVE-2026-88771: Citrix NetScaler Pre-Auth Command Injection Explained

  • CISA AA26-281A: How Chinese Government-Linked Actors Steal Sensitive Data

    CISA AA26-281A: How Chinese Government-Linked Actors Steal Sensitive Data

  • CVE-2026-94127 Explained: F5 BIG-IP APM Heap Overflow Attack

    CVE-2026-94127 Explained: F5 BIG-IP APM Heap Overflow Attack

  • How Click2Shell Exploits WordPress Theme Installation for RCE

    How Click2Shell Exploits WordPress Theme Installation for RCE