Hackers Use LLMjacking Attacks to Hijack Company AI and Run Up Huge Cloud Bills

Cybercriminals are stealing access keys and cloud credentials to hijack expensive artificial intelligence services, according to a report from Google's Threat Intelligence Group.

Google warns that this practice, known as LLMjacking, has grown steadily over the last couple of years. Hackers typically start by finding publicly exposed artificial intelligence endpoints or stealing application programming interface keys. They often use stolen cloud identity tokens, GitHub access tokens, and active session tokens to get an initial foothold inside a target organization.

Once inside, attackers work to gain higher-level permissions and identify targets for further compromise. The ultimate goal is to monetize and resource-hijack foundation models through token proxying, automated model streaming, and graphics processing unit hardware provisioning. This means victims pay for the artificial intelligence usage while someone else uses their subscription.

Unlike cryptojacking, where mining is slow and costs accumulate over time, LLMjacking can run up six-digit bills in minutes. Google noted that one threat group, known as UNC6993, accumulated over $100,000 in billing charges by generating additional keys and enabling services within victim cloud projects. Another threat actor subscribed victim accounts to pay-as-you-go agreements, selected premium foundation models, and launched high-volume inference streams while using a burn-and-rotate strategy to constantly create and delete short-lived service accounts.

The cybercrime ecosystem behind these attacks is maturing quickly, with a growing division of labor between groups that establish initial access and operators who manage downstream monetization. These monetization managers, referred to as LLM Mixers, operate routing gateways that resell unauthorized artificial intelligence access to underground subscribers.

Google expects threat actors to increasingly prioritize LLMjacking over other forms of resource theft. The tech giant urges organizations to increase visibility over their artificial intelligence platforms, treat unauthorized feature enablement as a critical security event, and enforce strict access controls so users and accounts only have the permissions they strictly need.

Summary

As companies adopt expensive artificial intelligence tools, security teams must monitor their cloud infrastructure closely. Hackers are shifting toward stealing artificial intelligence access credentials because they can quickly drain corporate finances through high-volume usage before defenses notice.