ARTEX Agentic AI Hacking Platform Fuels Autonomous Cybercrime Campaigns at Scale
Check Point Research details how ARTEX, an originally open-source Chinese AI-powered penetration testing platform, is being abused by cybercriminals to run largely autonomous attack campaigns. Researchers identified 2,557 systems hosting or having hosted ARTEX panels, with 1,608 live and publicly exposed instances, and obtained operator-console artifacts (including 10 misconfigured instances exposing full databases) revealing prompts, models used and attack outcomes. Most operations required only an initial goal from the human operator; agents then autonomously conducted reconnaissance, exploitation and validation, predominantly using Chinese LLMs (Deepseek Flash family, Kimi K3.1 Pro), with Anthropic's Opus 5.5 triggering guardrails that led actors to switch to more permissive Chinese models. The developer made ARTEX closed-source on October 8, 2026 after it was observed in attacks on South Korean banks, but forks remain publicly available. Similar frameworks (CyberStrike, Cairn, PentAGI) and custom tools were also observed in the wild.
Detalhes técnicos
Three documented campaigns illustrate ARTEX's use: (1) an operator ran two consoles for ~2 weeks to compromise 10 Chinese gray-market iOS code-signing shops via hardcoded admin passwords, missing API authentication, RCE via command injection and SSTI, IDOR, weak credentials, and Linux privilege escalation, harvesting 300+ .p12 certificates (3 validated as usable); (2) a 4-day crypto-draining campaign against 292 crypto payment domains and various gambling/OTC platforms found numerous RCEs and PII exposures but achieved zero financial gain, including a hallucinated 'successful' exploit and a real supply-chain compromise of an AWS S3 bucket hosting 714 iOS enterprise-signed IPA landing pages (restored before abuse completed); (3) a 5-day campaign against 13 Chinese gaming-account credential backend servers recovered 190,000+ credential sets for Honor of Kings, with ~3.3% (≈6,000) still valid, via unprotected plaintext credential files and hardcoded admin credentials. C&C infrastructure analysis shows live ARTEX servers concentrated in the US (54%), Hong Kong (18%) and China (6%), with version-tag clustering suggesting at least three distinct operator groups.
Remediação
Organizations should assume exposure to high-volume, automated attacks and should test and harden internet-facing systems against common weaknesses such as hardcoded credentials, missing API authentication, command injection/SSTI, IDOR, and overly permissive cloud storage permissions.
Linha do tempo
-
ARTEX Agentic AI Hacking Platform Fuels Autonomous Cybercrime Campaigns at Scale
ler a análise desta fonte ocultar
Check Point Research details how ARTEX, an originally open-source Chinese AI-powered penetration testing platform, is being abused by cybercriminals to run largely autonomous attack campaigns. Researchers identified 2,557 systems hosting or having hosted ARTEX panels, with 1,608 live and publicly exposed instances, and obtained operator-console artifacts (including 10 misconfigured instances exposing full databases) revealing prompts, models used and attack outcomes. Most operations required only an initial goal from the human operator; agents then autonomously conducted reconnaissance, exploitation and validation, predominantly using Chinese LLMs (Deepseek Flash family, Kimi K3.1 Pro), with Anthropic's Opus 5.5 triggering guardrails that led actors to switch to more permissive Chinese models. The developer made ARTEX closed-source on October 8, 2026 after it was observed in attacks on South Korean banks, but forks remain publicly available. Similar frameworks (CyberStrike, Cairn, PentAGI) and custom tools were also observed in the wild.
Technical Details
Three documented campaigns illustrate ARTEX's use: (1) an operator ran two consoles for ~2 weeks to compromise 10 Chinese gray-market iOS code-signing shops via hardcoded admin passwords, missing API authentication, RCE via command injection and SSTI, IDOR, weak credentials, and Linux privilege escalation, harvesting 300+ .p12 certificates (3 validated as usable); (2) a 4-day crypto-draining campaign against 292 crypto payment domains and various gambling/OTC platforms found numerous RCEs and PII exposures but achieved zero financial gain, including a hallucinated 'successful' exploit and a real supply-chain compromise of an AWS S3 bucket hosting 714 iOS enterprise-signed IPA landing pages (restored before abuse completed); (3) a 5-day campaign against 13 Chinese gaming-account credential backend servers recovered 190,000+ credential sets for Honor of Kings, with ~3.3% (≈6,000) still valid, via unprotected plaintext credential files and hardcoded admin credentials. C&C infrastructure analysis shows live ARTEX servers concentrated in the US (54%), Hong Kong (18%) and China (6%), with version-tag clustering suggesting at least three distinct operator groups.
Remediation
Organizations should assume exposure to high-volume, automated attacks and should test and harden internet-facing systems against common weaknesses such as hardcoded credentials, missing API authentication, command injection/SSTI, IDOR, and overly permissive cloud storage permissions.