Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableWired→ original

Hackers Using the AI Framework Artex Gain Network Access in Under 3 Hours

Artex is an open-source tool that connects AI models to hacking tools. A hacker types a plain-language instruction such as "get access to the site's admin panel," and the agents hunt for weaknesses on their own. The group Storm-0940 used it to gain privileged access to systems in under 3 hours, a job that used to take days by hand. The traces are loud, so a forgotten test server with a default password is at risk, while patching and network traffic monitoring really do help.

AI-processed from Wired; edited by Hamidun News
Hackers Using the AI Framework Artex Gain Network Access in Under 3 Hours
Source: Wired. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceWired ↗
Editor-in-Chief
Key metric #1Vapor Panda attacked more than 200 organizations in the US and Europe with Artex
Key metric #2Dosu received 80,000 requests from Artex in 90 minutes
Key metric #3GitGuardian counted 45 AI frameworks for attacks, not just Artex

Artex is an open-source tool that turns an AI model into an automated hacker. The attacker writes a task in plain words, for example "get access to the admin panel, database, or server of this site," and several AI agents then scan the target, look for weak spots, and break in. This work used to take a human days; now it takes hours. According to Microsoft, hackers from the group Storm-0940 gained privileged access to systems in under three hours. For a developer, the implication is simple: a forgotten test environment with a weak or default password can now be found and attacked at scale, with no human on the attacker's side.

Security researchers have spent the past year warning that AI agents would greatly speed up cybercriminals. Now it is happening in practice. Since roughly late August, banks, government agencies, and companies around the world have become targets of attacks run by automated frameworks built on large language models. The most prominent of them is Artex, and researchers call it a "point-and-shoot" tool.

How Artex works

The framework is written in Go. A developer using the alias Autumn-27 posted it on GitHub in July. Artex itself is not an AI model. It is essentially a wrapper plus a set of instructions: it connects to external models (both frontier closed ones and open ones) through APIs and points them at familiar penetration-testing tools.

It works like a small crew: one agent breaks the plan into steps, others carry them out, and the system presents the results at the end. Researcher Jacob Larsen, founder of the company Cypen, explains how Artex differs from its predecessors. Earlier tools could do either reconnaissance only or vulnerability checking only. Artex covers the whole cycle: initial network analysis, intrusion, lateral movement inside the system, and staging data for exfiltration.

There is also a worrying detail about the models. The safety restrictions of frontier labs can stop an intrusion when they detect it. However, free Chinese models and "abliterated" models, which have had their refusal mechanisms removed, are getting better at carrying out intrusions.

Who is attacking whom

There are no new hacking techniques here, and the tools are not sophisticated, according to about half a dozen security companies. What changes is speed and the barrier to entry. Guillaume Valadon of GitGuardian says the bar is so low that almost any beginner can manage it.

Examples from the reports:

  • According to CrowdStrike, in October the China-linked group Vapor Panda used Artex to attack more than 200 organizations: think tanks, semiconductor and defense companies in the US, and foreign ministries in Europe.
  • Microsoft observed Storm-0940 trying to enter Microsoft 365 environments after a supply chain attack.
  • Researchers link Artex to the theft of personal data from thousands of South Korean financial organizations, possible cryptocurrency theft, and GitHub searches for accidentally published credentials.
  • Last week the FBI issued a separate warning: Artex is being used against the financial, government, technology, defense, energy, healthcare, telecommunications, and legal sectors.

On October 8, the author deleted Artex from GitHub and said he was not connected to malicious attacks and had built the tool for education and research. It changed nothing: copies had already spread and continue to be developed. GitGuardian has documented 45 similar AI frameworks, among which researchers often mention CyberstrikeAI, Hexstrike, CAIRN, and PentAGI.

What to do in practice

The good news is that these tools are currently noisy. They fire off thousands of probes in a row: the AI coding platform Dosu reported that Artex sent 80,000 requests to its systems in 90 minutes. Decent monitoring will catch a spike like that. Silas Cutler of Censys stresses that such tools are not stealthy yet.

According to the researchers, basic measures close off many of the attackers' routes:

  • install patches promptly;
  • do not use default credentials;
  • do not leave secrets and keys publicly exposed;
  • watch for unusual network activity.

It is hard to estimate how many people use Artex. According to analysis by Censys and the Shadowserver Foundation, thousands of deployed instances of such systems are exposed on the internet. Some operators left their servers unprotected, and researchers were able to see how the tool is used. Cutler saw two such servers where Chinese-speaking operators attacked Chinese organizations. In one case, a university was attacked with password brute-forcing, which he says looks like a student trying to change grades. His conclusion: part of the activity is organized, experienced criminal work, and part is closer to teenage mischief.

Story

Key Questions Answered

How Artex works?

The framework is written in Go. A developer using the alias Autumn-27 posted it on GitHub in July. Artex itself is not an AI model. It is essentially a wrapper plus a set of instructions: it connects to external models (both frontier closed ones and open ones) through APIs and points them at familiar penetration-testing tools. It works like a small crew: one agent breaks the plan into steps, others carry them out, and the system presents the results at the end.

Who is attacking whom?

There are no new hacking techniques here, and the tools are not sophisticated, according to about half a dozen security companies. What changes is speed and the barrier to entry. Guillaume Valadon of GitGuardian says the bar is so low that almost any beginner can manage it. Examples from the reports: According to CrowdStrike, in October the China-linked group Vapor Panda used Artex to attack more than 200 organizations: think tanks, semiconductor and defense companies in the US, and foreign ministries in Europe.

What to do in practice?

The good news is that these tools are currently noisy. They fire off thousands of probes in a row: the AI coding platform Dosu reported that Artex sent 80,000 requests to its systems in 90 minutes. Decent monitoring will catch a spike like that. Silas Cutler of Censys stresses that such tools are not stealthy yet.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…