Google's Mandiant business has warned that cybercriminals are increasingly targeting large language model (LLM) resources in a practice it calls "LLMJacking." The company said it responded to multiple incidents this year across financial services, retail, technology, and transportation sectors.
Mandiant said LLMJacking can generate large spikes in fraudulent cloud charges within hours. In individual incidents, it observed fraudulent charges tied to resource consumption exceeding $100,000 in a single day. Cumulative charges at some organizations reached $1 million before access was revoked.
According to the report, attackers bypass front-end chatbots to reach backend model resources. They typically do this using LLM API keys or exposed or stolen cloud identity tokens, including IAM credentials, Service Account keys, GitHub Personal Access Tokens, and active session tokens. From there, they escalate privileges and conduct internal reconnaissance to find more credentials. The report also said attackers may target vulnerabilities in LLM routing software to escalate privileges.
The report describes the end goal as monetizing and hijacking foundation model resources through token proxying, automated model streaming, and GPU hardware provisioning. Mandiant said it has seen financially