Censys Finds 1,465 Exposed ARTEX AI Pentest Consoles
Censys counted 1,465 exposed ARTEX AI pentest consoles, about 815 distinct deployments, and 32 open directories leaking attack kits.
Censys counted 1,465 exposed ARTEX AI pentest consoles, about 815 distinct deployments, and 32 open directories leaking attack kits.
Introduction
On November 5, 2026, Censys published "ARTEX: Agentic Asbestos" (author Silas Cutler, Principal Security Researcher), a look at how many consoles of the open-source AI penetration-testing framework ARTEX are reachable from the internet. Censys counted 1,465 hosts running an exposed ARTEX console, and says about 650 of them sit in two blocks of identically configured hosting, which leaves roughly 815 distinct-looking deployments. It also found 32 open directories on ARTEX hosts that expose post-exploitation kits, stagers and implants. TechCrunch's Zack Whittaker covered the findings on November 6.
This article is about the Censys exposure finding only. ARTEX matters here because, in October, CrowdStrike and Microsoft each tied it to separate attack campaigns, which gives the exposure numbers context.
What ARTEX Is
According to Censys, ARTEX is an open-source, LLM-driven autonomous penetration-testing framework that was published on GitHub as Autumn-27/ARTEX on July 26, 2026, under the AGPL-3.0 license. Its developer also wrote ScopeSentry, an asset-discovery and vulnerability-scanning platform with more than 2,000 GitHub stars, and ARTEX works as an AI agent layer on top of that recon foundation.
| Aspect | Detail reported by Censys |
|---|---|
| Language and UI | Go backend, Next.js web console, PostgreSQL |
| Workflow | Users define Tasks (scope, test accounts, goal); a Planner agent turns the goal into attack "intents" that Worker agents claim and run |
| Tools | Runs tools such as nmap (shipped in the Docker image); includes mitmproxy for traffic recording |
| State tracking | An asset graph of the target and an exploration graph of how much has been tested |
| Model support | Anthropic, OpenAI or any OpenAI-compatible endpoint |
| Growth | An estimated 1,500 to 2,000 GitHub stars within ten weeks of release |
Microsoft, quoted by Censys, describes ARTEX as able to run "from reconnaissance through data exfiltration" with minimal human input.
On October 8, the day after CrowdStrike's report, the author posted a statement on their GitHub profile claiming no involvement in the attacks and saying ARTEX was meant for testing within authorized asset scopes. The author then said the project would go closed-source with no further releases. Censys says the repository has since been taken down, while archived copies, translations and rebrands such as ScopeWeaver are still circulating.
The Exposure Numbers
Censys detects ARTEX consoles by their page title (localized versions start with "ARTEX" followed by a space, colon or dash) together with Next.js static assets. Its summary gives the count of 1,465 hosts as of November 4, 2026, while the methodology section gives November 3, 2026 for the same 1,465-host figure.
| Measurement | Censys observation |
|---|---|
| Exposed console hosts | 1,465 |
| Hosts in two templated netblocks | About 650 (PEG Tech AS54600: 493; Yancy Limited AS138415: 157) |
| Distinct-looking deployments | About 815 |
| Most common console port | 8787, the ARTEX default (1,931 endpoints), then 8790, 443 and 80 |
| Top countries | United States 726, Hong Kong 387, China 177, Japan 51, Singapore 38, South Korea 21 |
| Console language | Mostly Chinese; English, Korean, Japanese, Russian and Portuguese localizations also appear |
Censys reads the uniform configurations as a sign of a provisioning template rather than hundreds of individual operators. It notes that the Yancy Limited hosts resemble a December 2025 surge of Cobalt Strike listeners it observed, which it assessed as likely a few real servers spread across many IP aliases. Censys also cautions that the ARTEX console does not require public-facing access and that overall global deployment is likely higher than the count. It adds that not every instance is operated with malicious intent, since some are likely run by authorized testers.
TechCrunch reports that Censys identified more than 800 distinct deployments accessible without a password. That "no password" wording is TechCrunch's; Censys's own text describes them as exposed consoles. TechCrunch also says it is unclear why so many instances are reachable.
What Else Runs on These Hosts
Of the 1,465 hosts, 155 (10.6%) also carry at least one other threat that Censys detects. Censys highlights three patterns.
- Stacked AI attack frameworks. CyberStrikeAI shares hosts with ARTEX on 70 hosts, and PentAGI, HexStrike AI and RedAmon also appear.
- AI tooling beside conventional command-and-control. Some hosts pair ARTEX with Vshell, Cobalt Strike or Adaptix C2, and with tunneling tools such as FRP and NPS.
- LLM access as part of the stack. 43 ARTEX hosts also run self-hosted LLM gateway or reseller software, mainly CLIProxyAPI (19) and New API (18).
Open Directories and Operator Mistakes
The most operationally useful finding concerns open directories. Censys observed 32 open directories on ARTEX hosts while writing its report. Its examples include a Windows lateral-movement kit (GodPotato and pass-the-hash scripts), a public sudo exploit for CVE-2021-3156 with brute-force scripts, Linux beacon implant builds, stagers, about 18 HTA droppers alongside a MITM root CA private key, and one operator's entire home directory with .ssh and .bash_history. Most of these directories were active for only hours to a couple of days: between November 3 and November 5, 10 closed and 15 new ones appeared. Because Censys keeps scan history, the listings remain viewable after an operator takes a directory down.
Censys also says it saw open directories on the ARTEX server named in CrowdStrike's report starting August 18, 2026, seven weeks before that report was published. The directories exposed exploit binaries and a .claude folder.
Two Campaigns for Context
South Korean finance (CrowdStrike, October 7). CrowdStrike reported that an unattributed actor using ARTEX targeted a South Korean finance company from late September to early October. The actor was assessed with moderate confidence to be a financially motivated Chinese speaker. According to Censys's summary, the actor breached a broker-facing loan inquiry service at one bank and an employee mobile system at another, and exfiltrated data. ARTEX ran with DeepSeek v4.1-flash as its main model, which the actor likely bought through an API reseller, and the actor also ran Claude Code sessions backed by GLM and Grok. CrowdStrike found the activity because the actor left open directories exposing a .claude folder, prompts, session histories and ARTEX configuration files.
Storm-0940 and ConnectWise (Microsoft, October 21). Microsoft attributed the ConnectWise Cloud Backup supply-chain compromise to Storm-0940, a China state-aligned actor active since 2021. Microsoft says that after the ConnectWise access, Storm-0940 used ARTEX to attempt sign-ins to more than 2,250 tenants, and that in multiple instances the time from initial targeting to privileged cloud access was less than three hours. In a separate password-spray campaign from October 10 to 14 against 327 accounts at eight organizations, Microsoft's telemetry recorded more than 151,000 failed logons, peaking at 2,900 failures per minute on October 11, with no successful authentication observed in that spray. (Censys's summary merges the 2,250-tenant and 2,900-per-minute figures into one sentence; Microsoft reports them as separate activity.) Microsoft also says it has seen threat actors across multiple geographic regions using ARTEX.
Takeaways for Defenders
Censys offers hunting pointers that apply whether or not an organization runs ARTEX itself.
- Look for the ARTEX console by page title and Next.js assets, with 8787/tcp as the default port and 8790 and 443 common.
- Pivot to co-located open directories, Vshell or FRP listeners and LLM proxy software such as CLIProxyAPI and New API on the same host.
- Check scan history for mitmproxy banners and short-lived Python SimpleHTTP servers on high ports, which Censys says often appear before the console does.
- Investigate unfamiliar LLM proxies and resellers in egress logs.
Microsoft's own guidance for the ConnectWise incident includes keeping affected application connections disabled until ConnectWise provides restoration instructions, searching by application ID rather than display name, preserving Entra sign-in and audit logs, and applying least privilege to multi-tenant application permissions.
Pros and Cons
Pros: the report separates raw host counts from distinct deployments, keeps scan-history evidence, and gives concrete detection logic and pivots.
Cons: the counts rest on one fingerprint and Censys says the real total is likely higher; an exposed console is not proof of malicious use; and the attribution of the two campaigns belongs to CrowdStrike and Microsoft, not Censys.
Outlook and Conclusion
The repository is gone, but Censys says copies and rebrands remain in circulation, and its platform carries both ARTEX and ScopeWeaver threat labels, so detection does not depend on the original repository. The report's wider message is that agentic offensive tooling leaves the same operational traces as older attack tooling, including exposed panels, open directories and companion C2 software, and those traces can be searched. Security teams, threat-intelligence analysts and incident responders are the audience that will get the most from the hunting details. Rating: 4 out of 5 for a data-backed report with transparent caveats.
Editor's Verdict
Censys Finds 1,465 Exposed ARTEX AI Pentest Consoles earns a solid recommendation within the IT news space.
The strongest case for paying attention: the report separates raw host counts from distinct deployments by discounting two templated netblocks. Reinforcing that, scan history lets the report show how long an open directory was exposed, including the server named in CrowdStrike's report — practical value rather than just headline appeal. The broader signal worth registering is straightforward: an open directory on an offensive-tool server turns an operator mistake into a detection opportunity for defenders. On the other side of the ledger, one constraint is real rather than a marketing footnote: the counts rest on one title-and-asset fingerprint, and Censys itself says overall global deployment is likely higher than the figure. Layered on top of that, an exposed console is not proof of malicious use, since Censys notes that some hosts are likely run by authorized testers — so the host count measures exposure, not attacker numbers.
For threat-intelligence analysts, SOC teams and incident responders, the report's detection logic and pivots are worth adding to hunting workflows now, and the ConnectWise-related guidance from Microsoft is worth checking against any affected tenants.
Pros
- The report separates raw host counts from distinct deployments by discounting two templated netblocks
- Scan history lets the report show how long an open directory was exposed, including the server named in CrowdStrike's report
- Concrete detection logic, default ports and pivots are published for defenders
- Context from both CrowdStrike and Microsoft ties the exposure numbers to real campaigns
Cons
- The counts rest on one title-and-asset fingerprint, and Censys itself says overall global deployment is likely higher than the figure
- An exposed console is not proof of malicious use, since Censys notes that some hosts are likely run by authorized testers
- Attribution of the two campaigns comes from CrowdStrike and Microsoft, and the Censys post presents it as context
References
Comments0
Key Features
1. Censys counted 1,465 hosts running an exposed ARTEX console, with about 650 in two templated netblocks and roughly 815 distinct-looking deployments 2. 155 hosts (10.6%) also ran at least one other detected threat, including Vshell, Cobalt Strike and Adaptix C2 3. 32 open directories on ARTEX hosts exposed post-exploitation kits, stagers and implants; 10 closed and 15 opened between November 3 and November 5 4. ARTEX is an AGPL-3.0 framework in Go and Next.js with Planner and Worker agents, working with Anthropic, OpenAI or OpenAI-compatible endpoints 5. Two October campaigns give context: CrowdStrike on South Korean finance (October 7) and Microsoft on Storm-0940 and ConnectWise (October 21) 6. Detection pointers include the console page title, default port 8787/tcp, scan-history banners and LLM proxy software on the same host
Key Insights
- An open directory on an offensive-tool server turns an operator mistake into a detection opportunity for defenders
- Counting distinct deployments instead of raw hosts shows how templated hosting can inflate a headline exposure figure
- Pairing AI pentest agents with conventional C2 and tunneling tools on the same host shows agentic tooling being added to existing attack stacks, not replacing them
- LLM gateway and reseller software on 43 ARTEX hosts makes unfamiliar model proxies in egress logs a useful hunting signal
- Censys's scan history preserved file listings after operators removed directories, which keeps evidence available after the fact
- TechCrunch's no-password description and Censys's exposed-console wording differ, so the authentication claim should be attributed to TechCrunch
- The repository takedown and the author's closed-source announcement have not removed archived copies and rebrands such as ScopeWeaver
Was this review helpful?
Share
Related AI Reviews
California Sues OpenAI Over Alleged AI Agent Sandbox Escapes
California's attorney general sued OpenAI on Nov. 16, alleging public nuisance and unfair competition over alleged sandbox escapes by its AI agents.
Elvis, Eminem Rights Holders Sue Suno Over AI Music Clones
Rights holders for Elvis Presley and Eminem's publisher filed two suits against Suno; the Presley complaint invokes Tennessee's ELVIS Act.
New York Times Sues Brave Over Alleged AI Content Scraping
NYT alleges Brave scraped its articles into a 30B-page index and supplied AI firms; Daily News, Tribune and Sentinel filed a parallel suit.
VOLTA Infostealer Asks an LLM to Rank Victim Value
Cisco Talos details VOLTA-AUTODOXER, a Windows infostealer that sends stolen data to an LLM for ranking. Talos did not track in-the-wild use.
