Check Point says criminals are running near-autonomous campaigns with the ARTEX agent framework
Check Point Research says single operators used the ARTEX agent framework to steal certificates and game credentials for a few hundred dollars.
What Check Point found
On 5 November, Check Point Research published an analysis of how criminals are using ARTEX, an AI penetration-testing platform from China that was built for authorised security testing. According to the researchers, ARTEX uses large language models to coordinate several agents. These agents plan tests, run security tools and record evidence of weaknesses. The developer made the project closed-source on 8 October, but Check Point says forks and backups are still public and threat actors are using them.
The researchers said 2,557 systems had hosted ARTEX control panels. Of these, 1,608 were live and publicly reachable at the time of the research, and 10 exposed their full ARTEX database. Most live servers were in the US (882), followed by Hong Kong (295) and China (103). Check Point said Chinese models were the most widely used, with DeepSeek Flash the most common.
Three campaigns
Check Point described three campaigns, each run by a single operator:
- Code-signing certificates. Between 14 and 29 October, one operator ran two ARTEX consoles against 10 grey-market iOS code-signing shops, using DeepSeek v4.1 Flash and Kimi K3.1 Pro. Check Point says the operator took more than 300 certificate files and their passwords, but only three passed validation. It estimates the cost at US$700 to US$1,500.
- Cryptocurrency. A four-day campaign using DeepSeek v4 Pro first targeted 292 crypto payment domains. It then moved on to gambling, over-the-counter trading and custody services, and to one TRON wallet that the researchers say held about US$600,000. Check Point says no funds were moved. The agents found remote code execution flaws and database dumps, and one leaked cloud storage credential gave write access to a service's app bucket, but the operator did not exploit it.
- Gaming accounts. Over about five days, agents targeted credential servers at 13 Chinese mobile-game account shops. They took more than 190,000 credential sets from one site, where plaintext credentials sat unprotected in a web-accessible file. Check Point's sampling suggests about 3.3 per cent were still valid. It estimates the cost at US$180 to US$360.
Check Point describes the campaigns as opportunistic, financially motivated and nearly fully autonomous once the operator had set a goal. The cost figures are the researchers' own estimates based on token use.
Part of a wider pattern
This is the second report about criminal use of ARTEX in a week. On 2 November, BleepingComputer reported that CrowdStrike had been tracking a China-linked group it calls Vapor Panda. Since 17 October, the group has used ARTEX against more than 200 organisations in 14 countries and compromised at least nine. Targets included telecommunications providers and government bodies. In that campaign, BleepingComputer reported, the operators gave instructions in Chinese and stepped in at key decision points. The same article cited the Dutch Institute for Vulnerability Disclosure, which said an ARTEX-driven agent went from initial access to root on its network in 12 minutes.
The campaigns Check Point describes were less sophisticated and only partly successful. Three of more than 300 stolen certificates passed validation, and the cryptocurrency operation moved no money. Still, the figures suggest one person can now run an intrusion campaign lasting several weeks for a few hundred to about fifteen hundred US dollars in model usage. Several of the weaknesses the agents exploited were basic, such as credentials stored in plain text.
What is not known
Check Point's report does not name the victims, which were themselves grey-market businesses. It also does not say how many people are behind the 1,608 live panels. In the pages reviewed for this article, neither ARTEX's developer nor the providers of the models involved had commented publicly.
Sources
- Check Point Research: Hacking on autopilot, inside ARTEX cybercrime campaigns
- BleepingComputer: Chinese hackers use ARTEX AI in automated attacks against 200 orgs