The original
- Autumn-27/ARTEX, first commit 2026-07-26
- Won the Baidu Agent+ challenge on 2026-09-03
- Deleted on 2026-10-08 after the Korean bank breach reports
- Last open release: v0.3.15, commit e6ec569, AGPL-3.0
ARTEX.sh
An independent guide to the AI pentest framework: how it works, the 42 public copies, the benchmarks and the news.
- notable public copies
- 42
- lines of code in v0.3.25
- 173,504
- Baidu Agent+ score
- 67.83
Reported on by
The original ARTEX repo was deleted on 2026-10-08.42 public forks and copies are still online.
Today
- RuoJi6/ARTEX ships v0.3.25 with Docker images
- Translations in English, Korean, Chinese, Spanish and Japanese
- Archives keep the original 458-commit history
- A commercial edition, ARTEX Pro, sells on ThreatBook RedAgents
How ARTEX works:AI agents plan, run tools and prove findings in a loop.
01
Plan
A change in the graph wakes the planner. It reads coverage and scope, updates its todolist and dispatches intents, each tied to an asset.
02
Work
Three workers by default. Each claims one intent, runs real tools like nmap, curl and Playwright, writes facts and findings back, then stops.
03
Guard
Every tool call passes intercept rules, then an LLM judge, then a human. All traffic runs through a recording proxy on port 8788 as evidence.
04
Prove
Each write wakes the planner again until prove_goal closes the goal. A reporter merges duplicates, a retester checks fixes.
- 01A graph change wakes the planner (debounced, about 800 ms).
- 02The planner reads graph_overview, coverage and scope.
- 03It dispatches 0 to N intents into the frontier queue.
- 04A worker claims one intent and loads its assets.
- 05The worker runs tools through the guard and the recording proxy.
- 06It writes facts, assets and findings back to both graphs.
- 07The write wakes the planner again, until prove_goal.
Two graphs inside ARTEX:assets and findings, linked by anchors.
ARTEX is one Go binary with the Next.js console embedded, over PostgreSQL. It keeps what it knows in two graphs.
- tools exposed to the LLM
- 31
- PostgreSQL tables
- 60
- LLM wire formats
- 4
- workers per task by default
- 3
Exploration graph
One per task
- goal
- intent
- fact
- finding
- hint
Anchors link each intent, fact and finding to the assets it touched. That powers the coverage map.
- root domain
- subdomain
- ip
- service
- app
- endpoint
Asset graph
Global, shared by every task
42 public ARTEX forks and copies,sorted by type and stars.
Stars and last push from the GitHub API on 2026-11-02. Thousands of empty clones are left out.
See all 42 forksAll
42 of 42
- Maintained
- Translations
- Mirrors
- Archives
- Derivatives
- Offensive
- Rebuilds
- 01jiwoochris/artex-global2,521stars
v0.3.15last push 2026-10-14
- 02trhhubsd/ARTEX1,274stars
v0.3.15last push 2026-10-19
- 03Hinln/ARTEX1,239stars
v0.3.15last push 2026-10-08
Two public ARTEX scores:67.83 on Baidu Agent+, 89.78 on TSecBench.
Baidu Agent+ live range
2026-08-16
67.83/100
7th place, 5h 08m, 456M tokens
Tencent TSecBench
2026-09-03
89.78/100
6th place, 5h 03m, 271M tokens
National champion, 1st overall of 150+ teams in the first Baidu BSRC Agent+ challenge.
ARTEX news timeline:97 days, first commit to Google's report.
Every item is dated and linked to its source, in English, Korean, Chinese and Japanese.
Full timelineLanguage: EN
ARTEX and Friends: Enter Agentic Offensive Frameworks
Google Threat Intelligence Group + Mandiant
Language: EN
Former pentester used ARTEX AI to steal 8.2 million data records
BleepingComputer
Language: EN
China-Aligned Hackers Used AI Tool To Target Over 2,250 Cloud Environments Worldwide
Republic World
Language: EN
Microsoft: Chinese State Actor Used AI to Compromise ConnectWise Tenants
Infosecurity Magazine
Language: EN
Microsoft ties Storm-0940 to ARTEX-driven ConnectWise Cloud Backup compromise
Microsoft Threat Intelligence (via The Hacker News, 2026-10-22)
Language: JA
日本企業への不正アクセス、中国のAIツール「ARTEX」使用を確認
Impress Watch
Questions people search for
- What is ARTEX?
- An open-source AI system for autonomous penetration testing by Autumn-27: LLM planner and worker agents over a dual graph, written in Go with a Next.js console. AGPL-3.0.
- Where is the original ARTEX GitHub repo?
- Autumn-27/ARTEX was deleted on 2026-10-08 and returns 404. The last open release, v0.3.15, survives in archives such as Hinln/ARTEX.
- Which ARTEX fork is maintained?
- RuoJi6/ARTEX is the most active successor, with releases up to v0.3.25 and Docker images at ruoji6/artex.
- What is ARTEX Pro?
- A commercial edition sold by ThreatBook on RedAgents. It has no public source code.
- Was ARTEX used in the Korean bank breaches?
- CrowdStrike, AhnLab and Korean press tied ARTEX to the October 2026 breaches at South Korean financial firms. The author denies involvement.
- How does ARTEX work?
- A planner agent reads the target graph and dispatches tasks. Three worker agents run real tools like nmap, curl and Playwright, and a guard checks every tool call before it runs.
- What is the latest ARTEX version?
- RuoJi6/ARTEX ships v0.3.25. The last release from the original repo is v0.3.15, commit e6ec569, which most mirrors and archives keep.
- What license is ARTEX under?
- AGPL-3.0. Forks that run ARTEX as a network service must publish their source code under the same license.
- Which ports does ARTEX use?
- The web console and API listen on port 8787. The recording proxy that captures all tool traffic uses port 8788 with its own certificate authority.
- Is ARTEX available in English?
- Yes. The original interface is Chinese. jiwoochris/artex-global defaults to English, and other forks add Korean, Chinese, Spanish and Japanese.
- Is it safe to expose an ARTEX console online?
- No. v0.3.14 has a login bypass that lets an outsider take over the console. Upgrade, keep the console off the internet and put it behind a VPN.
- Is artex.sh official?
- No. It is an independent guide. It is not affiliated with Autumn-27, ThreatBook or any fork maintainer, and it hosts no code.
Follow the ARTEX story.
New forks, releases and reports, dated and sourced.
