Chinese Hackers Exploit Stolen LLM API Keys for Credential Theft via ARTEX

• •Chinese threat actors exploited stolen LLM API keys using ARTEX for credential theft.
• •Targets included multiple AI gateways, focusing on management interfaces for upstream credentials.
• •Known vulnerabilities CVE-2025-3248 and CVE-2026-28797 were exploited during the attacks.

Chinese-nexus threat actors utilized ARTEX, an open-source penetration-testing framework, to exploit stolen large language model API keys for credential theft. Amazon Threat Intelligence reported that these operations were detected between October 3 and November 5, 2026, through their MadPot decoy network. The attackers employed a self-sustaining inference loop, where stolen API keys financed further credential theft without needing additional computing resources. Targets included various AI gateways such as One API and LiteLLM, focusing on management interfaces that hold upstream model-provider credentials. The campaign also involved searching for secrets in client-side JavaScript and exploiting known vulnerabilities, including CVE-2025-3248 and CVE-2026-28797. Additionally, a separate campaign targeted cryptocurrency exchanges using deepfake technology against KYC checks. Amazon assessed the attribution to Chinese actors with moderate confidence but could not independently verify all outcomes.