Chinese Operator Used ARTEX AI in Global Hacking Campaign

Zscaler ThreatLabz has outlined a global ARTEX AI hacking operation that operated from early August to early October 2026, targeting organisations across 20 countries and multiple industries. The campaign used a customised ARTEX penetration-testing toolkit running inside a Docker container on a dedicated work server, with a separate jump server for lateral movement.

The attacker combined open-source hacking tools—nmap, sqlmap, nuclei, httpx, jadx, apktool, frida-tools—with a PostgreSQL 16 database and a modified ARTEX build to automate scanning, vulnerability discovery, and data exfiltration. The operation employed three custom shell scripts and a multi-model workflow, assigning roles to DeepSeek for initial target discovery, GLM for review, and Grok for cross-checking outputs, all orchestrated via API gateways and local shims.

Context across prompts was preserved with a Python hook that attached project summaries from local markdown files to new prompts, and the logs indicate targets were treated as preauthorized to avoid human intervention.

Investigators describe a Chinese-speaking operator, with prompt logs showing Chinese language prompts, Chinese-time configurations, and translations of stolen schemas into Chinese. While no charges have been announced, the evidence points to a financial motive rather than state espionage.

The attacks reached 20 jurisdictions including Canada, China, the UK, the United States, South Korea, and the UAE, hitting dating sites, SMS gateways, crypto exchanges, gift-card portals, gambling sites, retailers, schools, and logistics firms. The operator reportedly sought bulk personal data (mobile numbers) and payment gateway API keys, and warned against triggering direct withdrawals.

Practical responses include auditing externally facing web APIs, patching SQL injection flaws, monitoring for bursts from nuclei/sqlmap/httpx, blocking known jump-server IPs, storing payment keys in hardware-backed secret managers, and deploying decoys and database query rate limits to curb automated AI-assisted exfiltration.