Threat Research · research.checkpoint.com

ARTEX AI Agents Ran Largely Autonomous Cybercrime Campaigns, Check Point Finds

Nov 5, 2026 · Original article ↗

Summary

Check Point Research analyzed 10 exposed ARTEX databases and found operators using largely autonomous AI agents in opportunistic campaigns, including certificate and game-account credential theft and an attempted crypto-service supply-chain compromise.

Key points

Researchers found 1,608 ARTEX panels live and publicly accessible; 10 exposed instances revealed operator prompts, configured models, and agent activity.
In an observed campaign against 10 gray-market signing shops, agents obtained more than 300 iOS certificates and passwords; three passed the operator’s validation checks.
In a gaming-account campaign, agents found plaintext credentials and recovered over 190,000 Honor of Kings account records; sampling indicated an estimated 3.3% remained valid.
A crypto-focused campaign did not transfer funds to the operator. In one case, agents obtained write access to a bucket serving iOS app materials, but the operator did not successfully exercise the planned compromise and the original manifest was later restored.
The examined operators mostly checked progress or adjusted goals, while agents carried out tasks with little further interaction. Chinese models, particularly the DeepSeek Flash family, were prevalent in the analyzed instances.
Observed access methods included hardcoded or weak credentials, missing authentication, command injection, server-side template injection, and insecure direct object references.

Article Details

Attack Vectors: ARTEX agents targeted exposed signing-shop and gaming-account sites using hardcoded or weak administrator credentials, missing API authentication, IDORs, command injection, and server-side template injection to gain access.
In one cryptocurrency-service campaign, an unauthenticated API exposed AWS S3 credentials, enabling write access to a bucket used to distribute iOS apps and landing pages. The operator's agent altered distribution content as part of a supply-chain attack.
Agents found plaintext gaming-account credentials in an unprotected web-root file and discovered administrator credentials in frontend JavaScript and HTML.
Operators configured residential proxies to work around WAF restrictions and IP blocks.
One campaign tasked agents with denial-of-service attacks against gambling sites; the article does not establish that those attacks succeeded.
Defensive Notes: Check Point Research advises testing and hardening systems in the exposed perimeter against high-volume agentic attacks.
The reported compromises involved exposed administrative interfaces, missing API authentication, weak or hardcoded credentials, vulnerable web applications, and overly permissive access to files and storage buckets.

Indicators of compromise

Type	Indicator	Context
IPV4	`103[.]122[.]245[.]194`	ARTEX C&C server IP in a cluster the researchers believe with high confidence is managed by the same operators.
IPV4	`104[.]168[.]112[.]237`	ARTEX C&C server IP in a cluster the researchers believe with high confidence is managed by the same operators.
IPV4	`192[.]236[.]172[.]197`	ARTEX C&C server IP in a cluster the researchers believe with high confidence is managed by the same operators.
IPV4	`198[.]44[.]123[.]66`	ARTEX C&C server IP in a cluster the researchers believe with high confidence is managed by the same operators.
IPV4	`204[.]44[.]66[.]81`	ARTEX C&C server IP in a cluster the researchers believe with high confidence is managed by the same operators.