Censys finds 1,465 exposed ARTEX AI hacking consoles, some beside exploit kits
One host held an operator's entire home directory, shell history and all.
Part of a running story: ARTEX, the hacking agent that keeps turning up (4 parts)
The internet-scanning company Censys says it found 1,465 hosts running an exposed console for ARTEX, the open-source AI hacking agent that has turned up in two big breaches this fall. The count is as of November 4, 2026, and it's in a report by Silas Cutler, a principal security researcher at Censys.
The headline number shrinks when you look closely. About 650 of those hosts sit in two blocks of identically configured hosting, which leaves roughly 815 deployments that look distinct. Censys also says the console doesn't need to face the public internet at all, so the real total is likely higher.
- Exposed ARTEX consoles
- 1,465 hosts, as of November 4, 2026
- Distinct-looking deployments
- Roughly 815
- Also running other threats
- 155 (10.6%)
- Open directories
- 32
- Most common port
- 8787, the ARTEX default
What was left in the open
The more useful finding, in Censys's own words, is how often operators leave files lying around. It saw 32 open directories on ARTEX hosts while writing the report. One held a Windows lateral-movement kit, another a public exploit for the 2021 sudo bug known as Baron Samedit. A third exposed an operator's entire home directory, SSH folder and shell history included.
Most of those directories were up for hours or a couple of days. Between November 3 and November 5, 10 closed and 15 new ones appeared. Censys keeps scan history, so the file listings stay readable after an operator takes a directory down.
It also went back to the server CrowdStrike named in its October 7 report on the South Korean bank breaches. Censys saw open directories there starting August 18, 2026, seven weeks before that report came out, exposing exploit binaries and a .claude folder.
Who's running it
Of the 1,465 hosts, 155 also run at least one other threat Censys detects. The most common companion is another AI attack tool: CyberStrikeAI shares 70 hosts with ARTEX. A smaller set pairs ARTEX with Cobalt Strike, Vshell or Adaptix C2, the C2 tools attackers have used for years.
The two big hosting blocks belong to PEG Tech, with 493 hosts, and Yancy Limited, with 157. Censys reads those as templates, and it ties Yancy to a surge of Cobalt Strike listeners it saw in December 2025.
Uniform configurations like these suggest a provisioning template rather than hundreds of individual operators.
Not every console is hostile, and Censys says so: authorized testers run ARTEX too. The report doesn't try to split the 815 into criminals and testers. (I'd read the 70 hosts that also run CyberStrikeAI, and the ones next to Cobalt Strike, as the clearest cases, but that's our reading, not Censys's.)
What it means for the companies on the other end
ARTEX's targets so far have been ordinary businesses and their customers. CrowdStrike's actor breached a broker-facing loan inquiry service at one bank and an employee mobile system at another, and took data. Censys says the files on that server show work against payment and gaming services too.
Microsoft tied the tool to Storm-0940's ConnectWise supply-chain compromise, where the group tried sign-ins to more than 2,250 customer tenants, at up to 2,900 attempts a minute. Some victims saw privileged cloud access in under three hours.
Taking the code offline hasn't taken the tool offline. Its author, Autumn-27, said on October 8 that the project would go closed-source, and the repository is down, but archived copies, translations and rebrands such as ScopeWeaver still circulate. The largest group of exposed consoles is in the United States, with 726 hosts, ahead of Hong Kong with 387 and China with 177.
For defenders, Censys points at the plumbing. 43 ARTEX hosts also run self-hosted LLM gateway or reseller software, and it suggests checking egress logs for unfamiliar LLM proxies. Its default-port tip is 8787/tcp, with 8790 and 443 also common.
More on Security
- US prosecutors charge three men with using AI-cloned CEO voices to take $129 millionNovember 7, 2026
- OpenAI rates GPT-6 Cyber Critical in cybersecurity and limits it to vetted Daybreak Red usersNovember 5, 2026
- UK cyber agency says AI-powered ransomware on British infrastructure is highly likely by 2028November 4, 2026
- Sam Altman tells Stanford students AI is finding bugs faster than patching can fix themOctober 25, 2026