The Pentest Tool That Went Rogue: What the FBI’s ARTEX Warning Really Says

Amigos, Friday afternoon the FBI dropped a FLASH that a lot of you probably saw scroll by on X: FLASH-20261113-001, “Threat Overview of ARTEX Autonomous Penetration Testing Framework,” dated November 13, 2026, and marked TLP:CLEAR.

The headline line is a good one. In the FBI’s words, threat actors “have weaponized” ARTEX “to automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration.”

Machine speed. That’s the FBI’s phrase, not mine, and I’m quoting it on purpose. Nobody measured it in a lab for this post. But it’s the right phrase to sit with, because it’s the whole story.

What ARTEX actually is

First thing, por favor: ARTEX is not malware. It’s an open source penetration testing framework that got abused.

The FLASH describes it as “an autonomous, multi-agent penetration testing platform written in Go.” You give it goals in plain language, and it “translates high-level tactical goals described in natural language into actionable machine commands executed by specialized super intelligence (SI) agents, requiring minimal human interaction beyond initial objective setting.” SI is the FBI’s term, super intelligence.

It doesn’t ship its own model. It plugs into outside LLM providers: the FLASH lists Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok and Kimi. In the attacker instances analyzed, the FBI says Chinese models were most common, with DeepSeek Flash family models dominating because of capability, low cost, and “lack of observable guardrails when framed for penetration testing.”

Think of it like a really good power tool. A contractor uses it to build your deck. A burglar uses the same tool to get through your back door faster. The tool didn’t change. The hands did.

The developer is not the bad guy

The original project went up on GitHub in July 2026 under the handle Autumn-27. Censys dates the repo to July 26, under the AGPL-3.0 license. Per the FLASH, it picked up an estimated 1,500 to 2,000 stars in about 10 weeks.

On October 8, after reports of abuse, the developer pulled the repository. The statement still on the Autumn-27 GitHub profile (in Chinese) says, in our translation, that the attacks are “unrelated to the tool’s author,” that ARTEX was meant for learning, research and authorized testing, that unauthorized use is strongly condemned, and that the project will stop updates and go closed source.

So Autumn-27 is a developer handle, not a threat actor. The FLASH says the same thing about intent. And the takedown didn’t stop anything: the FBI says ARTEX “continues to proliferate through community forks, localized variations, archived copies, and a commercial version (ARTEX PRO).” Once code is out, mira, it’s out.

Who’s using it, and what they’ve done

Here’s where people get sloppy, so let’s be careful about who says what.

The FLASH says “a variety of malicious cyber actors” have used ARTEX “since at least August 2026,” against organizations in financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications and legal. Not one group. A variety.

It names two notable campaigns:

  • South Korean financial institutions. The FBI calls it financially motivated, with personal information exfiltrated from approximately 67,000 individuals. CrowdStrike and Zscaler both assess, at moderate confidence, a Chinese speaking, financially motivated actor. Nobody has named the group.
  • ConnectWise Cloud Backup. The FLASH calls it a supply chain compromise that “enabled access to more than 2,250 downstream tenants.” Microsoft’s October 21 report, which the FLASH cites for that number, words it differently: the actor “used ARTEX to attempt to sign-in to more than 2,250 tenants.” Attempted sign in and enabled access are not the same claim, so I’m giving you both.

On that ConnectWise campaign, Microsoft assesses that Storm-0940, which it calls “a China-state aligned actor,” was responsible. That’s Microsoft’s attribution. The FLASH cites Microsoft’s paper but doesn’t use the Storm-0940 name or call it a nation-state campaign in its body. The FBI didn’t attribute this to China. Microsoft did.

The FBI also says some intrusions went from initial targeting to privileged access in under three hours, citing Microsoft, which says “less than 3 hours” for several Storm-0940 attacks.

How many ARTEX consoles are sitting on the open internet? Depends who’s counting. The FLASH says over 1,600 as of early November, citing Censys. Censys’s own post says 1,465 as of November 4. Check Point says “more than 1,800.” Different measurements, different days. None of them is zero.

Why this matters

The danger here isn’t some new exploit. The FLASH lists no CVEs at all. The danger is tempo.

The FBI’s list of risks reads like a list of things defenders already struggle with: forgotten partner and supplier portals, mobile backends, legacy services, third party integrations, supply chain pivots, and a lower barrier for less skilled attackers. Then the kicker: a compressed detection window, because attack speed can outpace a coordinated response even when the traffic is noisy.

Translation: the attacker doesn’t need to be quiet anymore. They just need to be faster than your ticket queue.

For OT and ICS operators (our analysis, not the FBI’s)

Let me be straight with you, familia. The FLASH has no OT or ICS content. No SCADA, no PLCs, no industrial protocols. Energy and manufacturing show up only in the sector list. Nothing here describes an intrusion into a control network.

So this is my read, not a FLASH finding. What ARTEX goes after is the stuff around the plant: identity, cloud tenants, vendor portals, backup platforms, remote support paths. That’s the front porch of most OT environments. If an automated intruder owns your enterprise identity in under three hours, the question becomes how thick the wall is between that and your engineering workstations.

What I’d do this week:

  • Hunt on the enterprise side first. Use the official FLASH PDF and CSV, and vet every indicator against current telemetry before blocking. The FBI warns cloud IPs get reassigned.
  • Don’t block major LLM API domains as a standalone ARTEX signal. Those are legit services. Unexpected egress is a lead, not proof.
  • Inventory every internet facing app, especially vendor, backup and remote access paths. Get management interfaces off the public internet.
  • Keep control networks segmented from enterprise identity and cloud, per NIST SP 800-82r3 and CISA’s Cross-Sector Performance Goals.
  • If you run ConnectWise Cloud Backup, follow ConnectWise’s restoration guidance and review from at least October 1, 2026, as the FLASH recommends.
  • Report confirmed activity to your local FBI field office or IC3.

Don’t wait for an “ARTEX ICS signature.” There isn’t one.

Bottom line

  • ARTEX is a legitimate open source pentest framework that “a variety of malicious cyber actors” have abused. The developer is not the attacker.
  • The FBI’s warning is about speed, not a new vulnerability. Attribution to Storm-0940 is Microsoft’s, not the FBI’s.
  • For OT, the risk is the IT, identity and vendor surface around the plant. Shrink it and keep the wall between enterprise and control networks thick.

Stay sharp out there.

Javier

Sources

  1. https://www.ic3.gov/CSA/2026/261113.pdf
  2. https://www.ic3.gov/CSA/2026/261113.csv
  3. https://github.com/Autumn-27
  4. https://www.microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise/
  5. https://censys.com/blog/artex-agentic-asbestos/
  6. https://research.checkpoint.com/2026/hacking-on-autopilot-inside-artex-cybercrime-campaigns/
  7. https://csrc.nist.gov/pubs/sp/800/82/r3/final
  8. https://www.cisa.gov/cross-sector-cybersecurity-performance-goals

Leave a Reply