top of page

ARTEX: Notes on an Open-Source Agent Framework the FBI Flagged

2m
4 min read
ARTEX: Notes on an Open-Source Agent Framework the FBI Flagged
ARTEX: Notes on an Open-Source Agent Framework the FBI Flagged

On November 13, the FBI's Internet Crime Complaint Center published FLASH-20261113-001, a TLP:CLEAR advisory devoted to a single piece of software. ARTEX is an autonomous, multi-agent penetration testing platform written in Go, and according to the bureau, a variety of malicious actors have used it since at least August 2026 against organizations in financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications, and legal. Citing Microsoft, the FBI says some intrusions went from initial targeting to privileged access in under three hours. The full advisory, with indicators of compromise and mitigations, is posted at ic3.gov.


A legitimate tool with no model of its own

A developer using the GitHub handle Autumn-27 published ARTEX in July 2026. The Wall Street Journal later identified the developer as Li Puhua, a Chinese cybersecurity engineer. The project carried an AGPL-3.0 license and instructions to run it only against systems the operator owned or was authorized to test.


A planner agent converts a written objective into attack "intents," and worker agents run tools against the target in a loop until the goal is met. ARTEX ships without a language model. Operators connect whatever provider they choose, and the FBI lists Anthropic's Claude, OpenAI's GPT, DeepSeek, GLM, Grok, and Kimi. In the attacker-controlled instances the bureau examined, Chinese models were most common, with DeepSeek's Flash family dominating. The FBI attributes that to cost and capability, and to what it describes as a lack of observable guardrails when requests are framed as penetration testing.


Three campaigns in the record

The advisory draws on three cases documented by private-sector researchers. The first hit South Korean financial institutions, where personal information on roughly 67,000 people was exfiltrated. CrowdStrike and Zscaler each assessed with moderate confidence that the operator was a financially motivated Chinese speaker, and CrowdStrike reported that the attacks paired ARTEX with Anthropic's Claude Code.


The second was the October compromise of ConnectWise Cloud Backup. Microsoft attributes it to Storm-0940, a group it describes as China-state aligned and active since 2021. After password spraying against 327 accounts produced no logins, the group switched to stolen ConnectWise application identities and used ARTEX to attempt sign-ins across more than 2,250 Microsoft 365 tenants, at a rate Microsoft measured at 2,900 attempts per minute. The third case involves a Microsoft-tracked cluster called Storm-3358, whose command-and-control and ARTEX servers were still active as of November 6.


Other vendors have added detail. CrowdStrike says a China-linked group it calls VAPOR PANDA began using ARTEX on October 17 and pointed it at more than 200 organizations in at least 14 countries in under two weeks. Amazon's decoy sensors captured ARTEX agents between October 3 and November 5 running a credential-theft operation in which stolen AI keys paid for the inference used to hunt for more keys. SentinelOne estimated each campaign cost between $145 and $1,500 in model time.


Removing the repository did not remove the tool

On October 8, after the South Korean attacks were reported, Autumn-27 posted that the project would go closed source with no further public releases or maintenance. The developer said ARTEX was meant to help organizations test their own security and disclaimed responsibility for illegal use. The repository came down. The FBI says the framework still circulates through community forks, localized variants, archived copies, and a commercial edition called ARTEX PRO.


Researchers have also counted the consoles. Censys found 1,465 hosts with exposed ARTEX consoles on November 4, about 815 of which looked like distinct deployments, and 155 of those hosts also ran tooling it flags as malicious, including Cobalt Strike and Vshell. Check Point counted 2,557 systems that host or had hosted ARTEX panels, 1,608 of them live. Silas Cutler, the Censys researcher who wrote its report, told TechCrunch that some exposed instances belong to authorized testers, and that the software has no need to face the internet at all.


What the FBI asks defenders to do

Many of the intrusions relied on ordinary weaknesses. TechCrunch reported weak or known passwords and missing multi-factor authentication across ARTEX-linked attacks, and the FBI's initial-access list names misconfigurations and common web flaws such as SSRF, SQL injection, and hardcoded passwords in JavaScript, with no specific CVEs cited. The bureau says the agents favor systems that get less security attention: partner portals, mobile backends, employee apps, and legacy services.


Its mitigations follow from that. The FBI recommends inventorying every internet-facing application, restricting administrative and Kudu interfaces to trusted networks, rate-limiting high-impact APIs, scanning code and pipelines for embedded secrets, and moving enterprise applications behind zero-trust access where feasible.


For detection, it points to Zscaler's guidance on unexpected outbound calls to LLM APIs such as api.deepseek.com and api.anthropic.com. Organizations that installed affected ConnectWise applications are told to review activity back to at least October 1 and keep those applications disabled until ConnectWise issues formal restoration instructions. The FBI is asking victims to report to IC3 or a local field office, including LLM billing anomalies and time-to-containment data.

About the Author

David Borish is the author of The Tony Hawk Paradox: When Video Games Predict Reality tonyhawkparadox.com and writes about AI and technology at davidborish.com


Tony Hawk Paradox Book
 
 

JOIN THE AI SPECTATOR MAILING LIST

CONTACT

Contacting You About:

Thanks for submitting!

New York, NY           

Db  @DavidBorish.com           

  • LinkedIn
  • Instagram
  • Facebook
  • X
Back to top

© 2026 by David Borish IP, LLC, All Rights Reserved

bottom of page