Google warns LLMJacking could cost businesses up to $1m

Google’s Mandiant unit has warned that enterprise exposure to LLMJacking is increasing, with multiple incidents reported this year across financial services, retail, technology and transportation sectors. In its findings, threat actors bypass front-end chatbots to reach backend model resources, typically via LLM API keys or exposed or stolen cloud identity tokens such as IAM credentials, Service Account keys, GitHub Personal Access Tokens and active session tokens.

From there, attackers escalate privileges, perform internal reconnaissance and seek additional cloud credentials, potentially exploiting vulnerabilities in LLM routing software to broaden access. The ultimate aim is monetisation through resource hijacking of foundation models, achieved via token proxying, automated model streaming and provisioning of GPU hardware.

Google notes a diverse set of financially motivated clusters conducting LLMJacking across North America, South America and Asia Pacific, with a developing division of labour between those gaining initial access and those monetising downstream infrastructure, described as “LLM Mixers.” These operators proxy, aggregate and resell access to LLMs for criminal customers, a model deemed so profitable that Google expects cybercriminals to prioritise LLMJacking over cryptojacking going forward.

The research highlights the risk of enterprises expanding production deployments of higher‑tier models and agentic workflows, which could enable more automated reconnaissance and exploitation of exposed API tokens and model routing middleware, enabling rapid siphoning of enterprise quotas before compromises are detected. The message from Google is that the resales and illicit access markets will continue to grow, with victims left to bear the resulting cloud charges.