Google Warns of LLMjacking: $100,000 in Fraudulent Use in a Day

How the attackers get in and cash out

The attacks hit financial services, retail, technology and transport firms in North America, South America and Asia-Pacific. Attackers gained entry through credential theft and session hijacking, and also collected valid credentials exposed in public GitHub repositories.

Stolen API keys were then bundled into illicit proxy gateways, known as "LLM mixers," and access to models was resold on cybercrime markets at steep discounts. In one case, an illegal token-routing service offered buyers unlimited use of frontier models for about $60 a month, leaving the victim organization with roughly $1 million in model usage charges.

In another case, the attacker used a "burn and rotate" strategy, constantly creating and deleting short-lived service accounts to get around API call limits. The actor also checked the cloud provider's billing records regularly to monitor the damage.

What organizations should do

The report recommends threat hunting, taking inventory of unapproved "shadow LLM" use inside the organization, breaking API key permissions into finer-grained scopes, and automated secret scanning.

John Hultquist, chief analyst at GTIG, said attackers use LLMjacking to gain an edge over defenders in scale, using agentic tools to steal tokens without victims noticing. He urged companies to tightly control access to LLM resources, review usage closely, enable multi-factor authentication and manage account credentials carefully.