Swarm Chasing: Hunting ARTEX and Other Agentic Offensive Frameworks in the Wild
We found 47 ARTEX variants, 2,335 exposed offensive AI consoles and three undocumented offensive AI agent tools, using 24 detection and fingerprinting methods.
Lately, all we've been hearing about is AI attack tools. ARTEX, Cairn, PentAGI and CyberStrikeAI are open-source AI pentest tools, and all four have now turned up in real attacks. ARTEX was used to steal data from South Korean financial organizations20 and against AWS accounts.27 Storm-0940, a China-state-aligned group, used it after compromising a ConnectWise cloud backup app,21 and VAPOR PANDA used it against foreign ministries and telecoms.23 A China-nexus group ran a customized fork of Cairn against at least 19 organizations,28 Google saw PentAGI aimed at the aerospace sector,24 and Kaspersky found CyberStrikeAI in attacks on education organizations.25
Every one of these tools is run from a web console, and operators often leave that console open on the internet. So we asked: how many are out there, and what's actually running behind them?
To find out, we read each tool's source code in an isolated sandbox, pulled out fingerprints that survive renaming, and searched public scan data.
- PentAGI398includes the 116 cloned consoles
- ARTEX family202steepest climb, from mid-October
- Agent Zero46general-purpose agent
- CyberStrikeAI15few certificates before mid-September in our data
- RedAmon5graph recon
How to read this: each line counts consoles whose TLS certificate is dated on or before that day, so it only covers consoles that serve TLS and undercounts every tool. Whoever creates a certificate sets its date, and CyberStrikeAI makes a new one each time it restarts, so read the shape, not the exact days. Dating consoles by certificate follows Houdin.io’s CyberStrikeAI timeline (September 29), and Check Point published an ARTEX timeline on November 5. ARTEX and PentAGI census counts are in the figure further down.
What we found
Most of the 2,335 consoles are stock installs. The rest include 436 ARTEX consoles in one hosting network, 116 copies of one PentAGI setup, and 179 renamed or forked ARTEX consoles that a search for the default title misses.
Only one console in our census links directly to the attacks above: 204.194.51.54, a stock ARTEX console that AhnLab ASEC listed as an attack IP in October. We didn't tie any other console to those campaigns.
The numbers kept climbing while we worked. Between our November 6 census and a rerun on November 18, Shodan results for the stock ARTEX title went from 1,483 to 2,085, up 41% in 12 days, and PentAGI went from 520 to 626. Some of that will be Shodan crawling more.
The ARTEX family tree
ARTEX's original repo was deleted on October 8. Its forks and builds kept running. Cloudflare mapped 204 public ARTEX forks on November 13; we wanted to see which ones were actually deployed, and what else was running that no public repo explains.
Autumn-27/ARTEXHTTP 404 · deletedFirst commit ≈26 Jul 2026, deleted 8 Oct. The copies kept going.
- English & global
- ARTEX (English)horizonl · 130+ forks of its own
- artex-globaljiwoochris + forks · 0.3.15-global.1
- Korean
- artex-kobear2u · ko, zh, en, es titles
- artex-koTomatamatom1
- Chinese derivatives
- 月落 ycodeadds WAF bypass, leak hunting
- 谛听 DITINGstill artex-web inside
- SCA AuditErnket/CodeScan-Artex · code audit
- Upstream title kept
- reArtexwufufu770 · repackaged
- artex-cD4rk-3ch0 · adds guardrails
- artex failsafefailsafe-security · seen deployed in RU
- Trojanized
- Docker imagestomicdd + congdonglinux · backdoored · 7,946 pulls
- Fleet & ARTES
- sco-fix2348 services, one network
- artes.20261010“ARTES” rename
- 0.3.15-2-gc727eadsame ARTES family
- Operator custom
- heker.12 / .19HekerAI
- 0.5.0-customEnglish fleet
- Korean em-dash title2 consoles · edited Korean fork
- Named builds
- pentest-expert0.9.1 · ScopeWeaver-branded
- hotfix-seed-castTencent SG
- gateway-switchgoal-refresh-gateway-switch · artex-mx
- Dev & dated
- phase-e-20261115built 2 days before we saw it
- process-cleanup1.0.4 · “Team 执行节点”
- 20261105date-only version
- ffef5e2-dirtydev build in production
- 0.4.3past the last public 0.4.2
- 4+ consoles
- ScopeWeaver×65
- Autonomous Pentest Agent×11
- 渗透工作台×8
- TGSEC 社区专属版×4
- 2 consoles each
- HekerAI×2
- 月落×2
- Cybrion×2
- LMTex×2
- AntiFox×2
- NKW×2
- airsuperMAN×2
- 1 console each
- GOOGTEX
- ARTIFEX
- SHRIKE
- SKARN
- 猎杀者
- 海盗船
- ExploitFinder
- MiniHunter
- BoomInner
- Hopps
- ARGUS
- Soua
- mose
- artex-mx
- Team 执行节点
- 安全渗透智能体
- Disguised
- Security Console
- AegisLoop
Each row is a generation. Forks are public repos and Docker images, private builds carry version strings found in no public repo, and the bottom row is the names ARTEX consoles wore in our scans (×2 means two consoles). Those names are console titles; companies with the same name are unrelated. Seven of the 54 entries repeat a variant already shown, such as a build seen under a name, which leaves 47 distinct variants, 42 of them seen running.
The bottom row is the names ARTEX consoles wore in our scans. SOCRadar, PenTestIT and Profero reported some of them too. A few stood out:
- SCA Audit — 代码审计平台: a public code-audit derivative of ARTEX (Ernket/CodeScan-Artex), which one operator runs next to PentAGI on the subnet described below. We have no evidence the repo's author is involved.
- HekerAI: a custom build,
0.7.0-heker.12and.19, that appears in no public repo. - 月落 (Moonset): ycode, a public ARTEX derivative that adds GitHub leak hunting.
- 安全渗透智能体: a generic title ("security pentest agent") on a console Shodan identifies as ARTEX, so a title search never finds it.
Shodan's product data also carries ARTEX version strings, and many of them match no public release. We found fourteen private builds, seven of them new in our November 18 rerun. sco-fix2 is the fleet build below, first reported by Check Point. 0.9.1-pentest-expert.20261117 is a ScopeWeaver-branded build dated the day before we saw it. 0.4.3 is past the last public 0.4 release, so someone kept building after the repo came down. And ffef5e2-dirty is a development build running in production on Tencent Cloud.
Three undocumented offensive agent tools
A final sweep on November 18 looked for consoles describing themselves in red-team and attack terms. Four held up after we checked each one twice, and we show three here: we pulled the exact text from Shodan's stored copy of each page, confirmed it with a second search that returns only that console, and looked for the tool in web search (English and Chinese), GitHub and Sourcegraph (including forks), CSDN, Wayback, AlienVault OTX, GitGuardian's 45-tool catalog and TSecBench. None of them appears in any of those sources.
| Tool | What its own page advertises | Why we call it offensive | What we can't say |
|---|---|---|---|
| 红刃 (Red Blade) | "红队攻击 Agent" (red-team attack agent), with open endpoints /api/exploit, /api/scan, /api/recon and /api/chat, no login | It calls itself an attack agent and exposes an exploit API with no authentication | Who runs it |
| RedTeam Platform (红队安全测试平台) | Streams agent sessions and campaigns, chains findings into "attack chains", and has a form to "register real targets"; behind a login | It's a working campaign dashboard aimed at real targets | Who runs it, and whether that testing is authorized |
| 鲸鱼 (Whale) red-team pentest console | "LLM 驱动的鲸鱼红队渗透测试系统控制台" (LLM-driven Whale red-team pentest system console) | The same server presents a self-signed certificate impersonating www.apple.com (created November 11) and runs a SOCKS5 proxy, and its IP was already in ARTEX IOC feeds built from FBI and Talos indicators | The IP was known; the console on it is what's new |
Everything here is what the tools' own pages advertise. We didn't call any endpoint or log in to anything. Each is a single console, so treat them as first observations rather than tools in wide use. Shodan captured Red Blade on November 6 and Whale on November 3, and the RedTeam Platform was live on November 18. Red Blade's and Whale's certificates are from October and November, and any of these consoles could disappear. GreyNoise saw none of them scanning. To reproduce the two with no login, search Shodan for http.html:"红刃 · 红队攻击" and http.html:"LLM 驱动的鲸鱼"; each returns one console.
The same sweep turned up two weaker candidates we're holding: 海棠 (Haitang), a pentest portal known only by its login page, and a vulnerability-mining "command center" built on the public AutoCodex project. We dropped others after checking: Magician Hunter was an empty app shell, one "AI red team" console only reused a generic login template, and 鸾鸟 turned out to be LuaN1aoAgent, already in GitGuardian's catalog.
Ten more candidates
Before that sweep, searching console titles in Chinese, Korean, Russian and Turkish turned up AI pentest consoles for tools we couldn't find documented anywhere. Most named Chinese tools turned out to be documented already (DITING, Stopen, Dhunter, xArgus and others), so they're not shown here. What's left are candidates: deployed consoles of tools with no public source code, mostly one instance each, with weaker evidence than the three above. Some belong to vendors or companies, such as PaaP, Pentest Check and APTS; we include them because their consoles are exposed to the internet.
- 穿石 SeepZHPage code capturedAI pentest agent
- APTSZHPage code capturedAI automated penetration testing system
- PaaPTRPage code capturedAI-assisted pentest & red team
- MELLMAPRUPage code capturedPentest reconnaissance
- Pentest CheckRUPage code capturedDeep auto-pentest
- PTAgentZHRed-team pentest agent platform
- 破隟ZHAutomated pentest platform
- 天戟ZHAI pentest platform
- 玄戈 XUANGEZHAutomated pentest platform
- HVAZHAI pentest platform
Consoles running AI pentest tools we couldn’t find in any public repo, vendor page or published research. None has public source code, so these are candidates, not confirmed new tools. For five we captured page code showing a working app; the strongest is 穿石 Seep, with its own agent-chat API and per-task model routing. The other five are known from their console title only.
Disguised consoles
Some consoles hide behind a codename or another product's title. We found these with targeted searches, such as a tool's page code with its default titles excluded, so they're on top of the census count.
- DarkAI: a renamed CyberStrikeAI console on ByteDance's Volcano Engine cloud. The title says DarkAI, but the page carries CyberStrikeAI's
brand-wordmark__coreCSS class. ScanMalware named DarkAI among 57 renamed CyberStrikeAI consoles on November 7, and our marker finds it independently. - Security Console — 安全审计控制台: an ARTEX console presenting itself as a security audit tool, with an LLM gateway (new-api) on the same server. We found it through a login redirect we learned from urlquery reports (see Methodology).
- Bob Research Labs: a console titled "We're coming to root your server!", with IRC, MariaDB and open directory listings, which AhnLab ASEC reported first.
Tracking fleets
Hundreds of consoles running the same tool could be hundreds of different people. So we looked for consoles that provably came from one disk image or one private build. Following the swarmcha.se team (our title is a nod to them), we call these fleets. A swarm would mean agents coordinating across machines; these are copies of one tool, with the planner and worker agents coordinating inside each console. A fleet tells you where the software came from. Who runs it is a separate question: a hosting provider selling the same ready-made server to many customers would look identical from outside.
We used three kinds of evidence, from strongest to weakest:
- A shared TLS certificate. PentAGI makes its own certificate on first start, named
CN=localhost, so two separate installs never share one. Servers that present the exact same certificate also hold the same private key, so they were copied from one setup, either a cloned disk image or copied certificate files. A non-default certificate name makes that clearer still. - A private build string. A version string like
sco-fix2that matches no public release means the servers run the same private copy. Lots of them in one network points to one source, but it can't separate an operator from a provider's preinstalled image. - An identical page hash. Scanners hash each page a server returns. Matching hashes mean the same build, but stock releases share hashes too, so this only backs up the other two.
- One hosting network436AS54600; 348 on build sco-fix2
- Stock ARTEX1,171default title, many networks
- Renamed and forked179a title search misses them
- Login prompt only39found by the "ARTEX" login realm
- 116 copies, one certificate116same certificate and page hash
- Other PentAGI311independent installs
- Cairn48incl. renamed forks
- Other AI pentest tools35found by non-English title search; many already documented
How to read this: each square is one exposed console service. Darker blocks are the groups that share one origin: 436 ARTEX consoles inside one hosting network, 348 of them on one private build, and 116 PentAGI consoles sharing one certificate. The red band in the middle is the renamed and forked ARTEX builds, which a search for the default title misses. Counts come mostly from Shodan and are lower bounds; 14 servers run two tools side by side.
The sco-fix2 fleet: 348 services in one network
348 services on 254 IPs report the ARTEX build sco-fix2, which matches no public release, and every one sits in AS54600 (PEG TECH). How big the fleet is depends on what you count: its page hash went from 362 services on November 6 to 438 on November 18, while Shodan's sco-fix2 version matches fell from 425 to 402.
Others found this cluster first. RST flagged the network on November 2, Check Point named the build on November 5 (255 hosts), Gcore found 305 matching pages, and SOCRadar counted 242 on November 12 and found a Sub2API gateway on port 8790 on many of the servers. Our counts match theirs. It could be one operator's staging setup or a provider selling servers with ARTEX preinstalled; scan data alone can't tell which.
One PentAGI setup, 116 copies
116 PentAGI consoles present the exact same TLS certificate and serve the same page hash. The certificate's name was changed from PentAGI's default localhost to a specific server's IP address. Put together, that's strong evidence they all came from one setup, most likely a cloned disk image. The copies sit in AS398823, another network run by the hosting provider PEG TECH. Gcore reported this cluster on November 9 (101 endpoints), and RST counted about 90 there on November 2. By November 18, certificates naming the same server were on 158 consoles.
One operator, two tools
On one /24 subnet, seven ARTEX consoles (the SCA Audit build) share one certificate and six PentAGI consoles share another. Both certificates use the same custom naming pattern, O=CodeScan-Artex self-hosted and O=PentAGI self-hosted, both name the same server, and they were created a day apart. It's the closest thing to an operator fingerprint in our data: most likely one person set up both tools. GreyNoise recorded the subnet scanning in early November.
What this means
Exposed consoles are a cheap, public signal of who's deploying these tools and how. Here's what we learned:
- Titles don't hold up. 179 renamed or forked ARTEX consoles don't use the default title, and a title search for Cairn returns mostly coffee shops and holiday cottages. Page code, certificates and build strings held up much better.
- Copies are easy to spot. A shared certificate or a private build string turns hundreds of separate-looking consoles into a handful of setups.
- It moves fast. In the 12 days between our census and the rerun, the stock ARTEX title count rose 41%, seven new private builds appeared (one dated the day before we saw it), and three undocumented offensive agent tools turned up.
- Most of this is shared work. Many teams published counts, clusters and fingerprints before us. The catalog below credits each one and splits out what's ours.
This is the kind of work we love to do at Delta, our skunkworks: chase a question across scan data, source code and other people's research until the picture holds up. Stay tuned for more.
Methodology
We cloned the open-source repositories for ARTEX (from the BaaRai mirror), PentAGI, Cairn, CyberStrikeAI, scopeweaver, RedAmon, Xalgorix, Agent Zero, HexStrike AI and PentestAgent into an isolated AWS sandbox, and looked for the parts of each web UI a rebrand rarely touches: CSS animation names, theme scripts, asset paths and how certificates get made.
Titles turned out to be a poor fingerprint. On November 6, a Shodan search for title:Cairn returned 199 results, and about three quarters were coffee shops, holiday cottages, universities and other unrelated sites. The same day, a search for reasonBeaconPulse, a CSS animation name in Cairn's template, returned 44, all real Cairn consoles, including forks renamed to titles like Zero. Shodan's own Cairn product filter found 16. For ARTEX, title, page and version fingerprints together found 1,825 consoles; GitGuardian independently counted 1,428 with different fingerprints.
Scan engines also miss things. The swarmcha.se team used urlquery.net, a public URL-scanning service, to trace an agent fleet. We searched its public reports for the ARTEX console title and found consoles Shodan hadn't indexed. Those reports showed a redirect from / to /login on port 8091, which we turned into a Shodan query that surfaced the disguised Security Console.
Everything here comes from third-party scan data (Shodan, InternetDB, urlquery.net, urlscan.io, GreyNoise, DNS blocklists, certificate transparency logs) and source review in the sandbox. We never connected to a console. We also removed 157 rows from our raw census of 2,492 that turned out not to be consoles: unrelated pages matching a generic marker, businesses named Artex, and a static page explaining PentAGI's prompts.
An exposed console isn't proof of an attacker. Some of these will belong to authorized pentesters, students and researchers, and some may be honeypots set up to catch people hijacking consoles. Scan data has other limits too. Counts come mostly from Shodan, so they're lower bounds. Certificate dates are set by whoever makes the certificate. Shared certificates and builds show shared software, and anything we say about operators or motives is our best reading of that evidence.
We're not publishing IP addresses for most consoles. Indicators, the full host list and the 157 removed rows are available to researchers and defenders on request.
The fleets we found
| Fleet | Size | Evidence | Assessment | Confidence |
|---|---|---|---|---|
| sco-fix2 fleet | 348 services, 254 IPs, all in AS54600 | Private build string sco-fix2 (348); one page hash (362 services); Sub2API gateway on port 8790 (SOCRadar) | One operator's staging infrastructure or an "ARTEX preinstalled" hosting product. Network: RST (Nov 2); build string: Check Point (Nov 5); Sub2API: SOCRadar (Nov 12) | Shared origin: high. One operator: low |
| 116 PentAGI copies | 116 identical copies; 158 consoles by November 18 named the same server | One shared non-default certificate; one page hash | Most likely one cloned image; operator unknown. Cluster first reported by Gcore (Nov 9) and RST (Nov 2) | Shared origin: high. One operator: low |
| One operator, two tools | 13 consoles on one /24: 7 ARTEX, 6 PentAGI | Two certificates with the same custom naming pattern, naming the same server, created one day apart | Likely one operator; GreyNoise recorded scanning in early November | Shared origin: high. One operator: moderate |
| November 3 group | 5 consoles | One shared certificate dated November 3, in netblocks on Spamhaus's SBL and DROP lists | One cloned setup; no scanning seen in early November (GreyNoise) | Shared origin: high. One operator: low |
| Let's Encrypt group | 5 consoles | One shared Let's Encrypt certificate for a real domain instead of the default self-signed one | The same certificate and key on all five points to one setup; the domain's owner isn't known to be involved | Shared origin: high. One operator: moderate |
| PentAGI and adult sites | 8 servers | One shared PentAGI certificate; the same servers host six adult-content domains with an identical Cloudflare name-server pair | Shared image is solid; the name-server pair suggests, but doesn't prove, one account behind the domains | Shared image: high. Same account behind the domains: low |
The fingerprint catalog
Every technique we used. The first part is ours; the second is what others published first, and what it found for us.
| Technique | What it finds | Result | Evidence and credit |
|---|---|---|---|
| Our fingerprints and methods | |||
| Cairn CSS animation names | reasonBeaconPulse and related animation names in Cairn's page template; catches renamed forks | 44 services on November 6, against 16 from Shodan's product filter; forks Zero, CPU and WorkerAnt | From Cairn's source (oritera/Cairn), tested on Shodan. PenTestIT published a different Cairn style name (intentChipGlow) |
| CyberStrikeAI wordmark class | brand-wordmark__core in the login page | 336 services, including renamed builds such as DarkAI | From CyberStrikeAI's source (AIPentest/CyberStrikeAI), tested on Shodan |
| CyberStrikeAI stylesheet | c2.css referenced in served pages | 543 services | From CyberStrikeAI's source (AIPentest/CyberStrikeAI), tested on Shodan |
| DeepAudit HTML comment | The "Terminal Retro & Cassette Futurism Typography" comment in DeepAudit's page | 319 services on November 18; DeepAudit is outside our census counts | From DeepAudit's source; a candidate, since how many matches are real DeepAudit consoles is not yet measured |
| PaaP theme key | The paap-theme key in PaaP's page | 1 console | From the live page; a candidate with a single match so far |
| Login-header mining | Consoles that show nothing but a password prompt. Shodan can't search the login header, so we downloaded every port-8787 login response and read the headers ourselves | 39 services on 37 IPs naming the realm "ARTEX", out of 2,813 checked | Our method. The "ARTEX" realm itself was published by RST Cloud and MalwareMustDie |
| Title punctuation | Whether a Korean ARTEX console runs the public fork as shipped or an edited copy: the public fork's source deliberately uses a colon in the title, so an em-dash title means someone changed it | 2 consoles with an edited Korean title | From the bear2u/artex-ko source comment. RST listed both title forms, and PenTestIT traced titles to the fork's config file |
| New private build labels | ARTEX version strings in Shodan's product data that match no public release | 13 labels, including 0.7.0-heker.12, phase-e-20261115, 0.9.1-pentest-expert.20261117 and 1.0.4-process-cleanup.20261115 | The version-string method was published by Check Point (Nov 5); these labels are ours |
| Seven shared-certificate groups | Consoles that present the exact same certificate, so they share one origin | Groups of 8, 7, 6, 5, 5, 4 and 4 consoles, none in a published cluster | Certificate-reuse clustering was published by SOCRadar and Gcore, which also reported the 116-copy group |
| urlquery.net console search | Console URLs in urlquery's public scan reports that Shodan had not indexed | 2 ARTEX consoles, and a / to /login redirect we turned into a Shodan query that found the disguised Security Console | Using urlquery as a sensor follows swarmcha.se; searching it for consoles is ours |
| Non-English title sweep | Console titles with Chinese, Russian and Turkish pentest-agent keywords, to find tools English searches miss | 自动化渗透 16, 智能渗透 10 and AI渗透 17 services; the candidates in the undocumented-tools figure | Ours. Hunt.io and ScanMalware published ARTEX's titles in other languages |
| Cloudflare name-server pairs | Domains whose identical name-server pair suggests one Cloudflare account | Six adult-content domains on the same servers as the 8-console PentAGI group | The technique is known OSINT (James Motherway); this link is ours |
| Published by others, used here | |||
| Default titles | The <title> each tool ships with | Stock ARTEX, PentAGI and Cairn consoles | RST Cloud (Nov 2) |
| Shodan product filter | Shodan's own tag for a product | product:"Cairn" found 16 of 51 Cairn consoles | Shodan built-in |
| PentAGI favicon hash | The hash of PentAGI's default icon | 420 services on November 6 | RST Cloud (Nov 2) |
| PentAGI certificate issuer | The default "PentAGI CA" certificate, visible even behind a login | 243 services | RST Cloud (Nov 2); confirmed in PentAGI's source |
| ARTEX login realm | A 401 response naming the realm "ARTEX" | The 39 login-only consoles above | RST Cloud; MalwareMustDie. PenTestIT traces the realm to the ycode fork |
| CyberStrikeAI theme key | cyberstrike-theme in the theme script | 409 services | Houdin.io (Sep 29) |
| Identical page hashes | The same page hash across hosts means the same build | One page hash on 362 services in AS54600 | Gcore, SOCRadar and Profero |
| Page code without the default title | A tool's page code, excluding its stock titles, to find disguised consoles | Security Console, DarkAI | ScanMalware published similar exclusion queries |
| Agent drop-box inboxes | webhook.site inbox IDs decoded from agent payloads in urlquery reports | 4 inboxes, all already expired | swarmcha.se's method, reproduced |
| Activity and abuse checks | Whether console IPs are scanning (GreyNoise) or sit in abusive netblocks (Spamhaus) | Context for the fleet table; dated observations | Standard enrichment services |
| Behavior on the victim side | ArtexProbe accounts, artex-probe-* files, 2 to 25 second bursts and rotating webhook.site URLs (GTIG); a think-then-act request cadence (Unit 42); i04probe sign-up accounts (Okta) | Detection in your own logs | Google GTIG, Unit 42, Okta |
| ARTEX user agents | artex-readonly/1.0, ARTEX-worker/1.0 and others | Detection in your own logs | Datadog Security Labs (Nov 9) |
References
Exposed-console and infrastructure research
- Houdin.io. Mapping 200 still undetected Chinese AI attack platforms in the wild. September 29, 2026.
- AhnLab ASEC. The Threat of AI-Driven Autonomous Penetration Testing, as Examined by ARTEX. October 14, 2026.
- AhnLab ASEC. Analysis of the Infrastructure of ARTEX, a Chinese AI Penetration Testing Tool #4: The Birth of Copy Cat. October 14, 2026.
- swarmcha.se. We found a Tencent agent fleet. October 14, 2026.
- NorForge. Automated Subversion and Attempted Takeover of AI Pentesting Harnesses. October 26, 2026.
- Profero IRT. One Operator, Several AI Attack Agents: The Ultra Suite and ARTEX. October 31, 2026.
- RST Cloud. Tempo, not technique: hunting the AI harnesses behind today's intrusions. November 2, 2026.
- Hunt.io. ARTEX Exposed: Mapping the Infrastructure Behind the South Korean Bank Breaches. November 4, 2026.
- Profero IRT. Inside the AI Operator. November 4, 2026.
- Check Point Research. Hacking on Autopilot: Inside ARTEX Cybercrime Campaigns. November 5, 2026.
- Censys. ARTEX: Agentic Asbestos. November 5, 2026.
- ScanMalware. Agentic Hacking in the Wild: How We Track ARTEX, CyberStrikeAI and Their Rebrands. November 7, 2026.
- Gcore. We found 1,698 Internet-visible AI offensive-security endpoints. Here's what the data reveals. November 9, 2026.
- GitGuardian. Beyond ARTEX: 45 AI Pentesting Tools and How to Find Them. November 9, 2026.
- Palo Alto Networks Unit 42. ARTEX and AI-Powered Red Teaming Tools: What Defenders Need to Know. November 11, 2026.
- SOCRadar. ARTEX: 916 Exposed AI Pentest Consoles Mapped in the U.S.. November 12, 2026.
- Cloudflare. Adapting to AI penetration testing orchestrators: The Modern Security Model. November 13, 2026.
- Cisco Talos. ARTEX and beyond: AI-enabled penetration testing platforms. November 13, 2026.
- PenTestIT. AI Pentesting Frameworks: Abuse, Lineage, Fingerprints. November 14, 2026.
Threat intelligence
- CrowdStrike. Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance. October 7, 2026.
- Microsoft Threat Intelligence. AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise. October 21, 2026.
- Wiz. ARTEX in the Wild: Storm-0940 Expands Targeting of IT and Software Vendors. October 22, 2026.
- CrowdStrike. VAPOR PANDA Uses AI-Driven ARTEX to Scale a Global Espionage Campaign. October 29, 2026.
- Google Threat Intelligence Group and Mandiant. ARTEX and Friends: Enter Agentic Offensive Frameworks. October 30, 2026.
- Securelist. CyberStrikeAI in the wild: the use of AI in real-world attacks on organizations. November 2, 2026.
- Okta Threat Intelligence. Agentic attack tools used to accelerate reconnaissance of identity flows. November 6, 2026.
- Datadog Security Labs. On the hunt for ARTEX: understanding agentic attacks in AWS environments. November 9, 2026.
- Palo Alto Networks Unit 42. Attackers Use Autonomous AI to Target Semiconductor, Defense and Civil Society Sectors. November 12, 2026.
- Federal Bureau of Investigation. FLASH-20261113-001. November 13, 2026.
- AlienVault OTX. ARTEX infrastructure indicators (pulse 6afa566f28ebebfb22046701). November 16, 2026.
Detection content
- MalwareMustDie. MMD-072-2026 - ARTEX detection signature pack - from IOC w/playbook process. November 13, 2026.
Trojanized tools
- Cloud Security Alliance. Trojanized Offensive AI Tooling: The ARTEX Docker Backdoor. November 6, 2026.



