Censys counted 1,465 ARTEX AI hacking consoles open to the internet. Not all of them belong to criminals.
ARTEX is a tool that runs penetration tests with AI agents, released as open source in July. Censys counted 1,465 of its control panels online in early November, and TechCrunch reports the researcher behind the count saying many are used by authorized testers.
“As of November 4, 2026, Censys observed 1,465 hosts running an exposed ARTEX console.” That sentence is from a post by Silas Cutler on the Censys research blog, published November 5, and the rest of this story is mostly about how much weight it can carry.
ARTEX is a tool for running penetration tests with AI agents. Microsoft’s Security Blog describes it as a system of LLM-driven agents that can run a penetration test on their own, from reconnaissance through data exfiltration. Censys says the code went up on GitHub as Autumn-27/ARTEX on July 26, 2026, under the open-source AGPL-3.0 license. TechCrunch reports the developer has since announced a move to closed source (more in the limits note below).
Why 1,465 is not 1,465 attackers
Start with the count itself. Censys says about 650 of the hosts sit in two blocks of identically configured hosting, which leaves roughly 815 deployments that look distinct. Censys flags the Yancy Limited block as one of the two. TechCrunch, which covered the report on November 6, wrote “more than 800 distinct deployments.” Censys’s own arithmetic gives roughly 815, and I’m using that.
Then there’s the question of who runs them. TechCrunch reports Cutler saying that not every ARTEX instance is malicious and that many are used by authorized testers. TechCrunch also says the 800-plus deployments were “accessible from the web without needing a password to log in.” Censys covers password-spraying and brute-force tooling but doesn’t say the consoles need no login, so that access claim rests on TechCrunch alone. TechCrunch also says the software “does not require them to be accessible from the internet.” My read: the safe reading is that these consoles are reachable from the internet, and that some are probably legitimate.
Where the consoles are
Censys breaks the 1,465 down by country, and by two hosting networks that hold a big share of them.
| Country | Exposed hosts |
|---|---|
| United States | 726 |
| Hong Kong | 387 |
| China | 177 |
| Japan | 51 |
| Singapore | 38 |
| South Korea | 21 |
PEG Tech (AS54600) accounts for 493 hosts and Yancy Limited (AS138415) for 157. Together that’s the roughly 650, or about 44% of the total.
Censys also found that 155 of the hosts (10.6%) run at least one other threat it detects, including Vshell, Cobalt Strike and Adaptix C2. CyberStrikeAI runs alongside ARTEX on 70 hosts.
A timeline that starts before the alarm
Censys says open directories appeared on the server named in CrowdStrike’s report starting August 18, 2026. By Censys’s wording, that was seven weeks before that report came out. I did not read CrowdStrike’s report, so this is Censys’s account.
TechCrunch reports attacks on Korean financial firms in September and an attack on IT support provider ConnectWise in October. On October 21, Microsoft published its own account of Storm-0940, which it says “used ARTEX to attempt to sign-in to more than 2,250 tenants.” Microsoft says the sign-in activity peaked on October 11 at 2,900 failures per minute.
Those are separate reports from separate companies, and I haven’t seen anything that ties every exposed console to one campaign.
Limits: I haven't tested ARTEX or touched any of these hosts, and I did not independently verify any figure beyond what the three linked sources state. I did not read CrowdStrike's report, and the Cairn comparison below rests on our earlier post, which I did not re-check. The consoles, hosts and country figures are Censys's, while the tenant sign-in numbers are Microsoft's and the Korean and ConnectWise attacks are TechCrunch's reporting. Censys's summary dates the count November 4, while its measurement section says November 3, and I've used November 4.
TechCrunch also reports that ARTEX's developer "has since announced it would go closed-source, effectively limiting further releases for paid customers." That is TechCrunch's claim, not verified by me.
What I think
My read: the headline number is the least useful part. 1,465 hosts, about 815 deployments and an unknown count of active attackers are three different measurements, and only the first is something a scanner can see. A tool being easy to find tells you it spread. It doesn’t tell you what it did.
My read: the timeline is the better detail. Open directories on a server for seven weeks before a vendor report named it means outsiders could have seen the tool’s infrastructure well before the report. I can’t say how many other tools are in that position.
The Cairn case had the same feature: open dashboards let a researcher watch an AI-driven campaign as it ran. That post was about what the attacks cost. This one is about how visible the tooling is.
Sources
- Censys ARC: “ARTEX: Agentic Asbestos” by Silas Cutler (November 5, 2026)
- Microsoft Security Blog: AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise (October 21, 2026)
- TechCrunch: Hundreds of ARTEX AI hacking tools found publicly exposed to the web (November 6, 2026)