Google reports LLMjacking attacks driving cloud bills to $1 million

Google has warned that attackers are hijacking enterprise access to Large Language Models (LLMs) to drain cloud resources and services, creating fraudulently large computing charges.

In a report shared with The Hacker News, Google Threat Intelligence Group and Mandiant said LLMjacking is a form of resource hijacking aimed at foundation models, where criminals obtain API keys through phishing, leaks, misconfigurations or exposed client-side code to access a company’s LLM accounts while the legitimate owner covers the costs.

The joint assessment outlines a global footprint, with affected organisations in North America, South America and Asia Pacific across financial services, retail, technology and transportation sectors. In observed incidents, fraudulent cloud charges have exceeded $100,000 in a single day, with cumulative bills reaching as high as $1 million before access was revoked.

Initial access commonly involves compromised cloud identity tokens—such as IAM credentials, Service Account keys, GitHub PATs, and active session tokens. Case details include UNC6993, which allegedly breached a cloud instance via a misconfigured Service Account key and used it to access an LLM API, including a session hijacking page that duped users into pasting raw JSON payloads containing active tokens to bypass MFA.

The report also highlights use of stolen Personal Access Tokens to access private code repositories, and the exploitation of a suspected vulnerability in containerized cloud environments. The analysis notes a mature LLMjacking ecosystem, including LLM Mixers that proxy and resell unauthorized LLM access, with services such as Jet Night offering broad access for around $60 a month. CVE-2026-40217 is cited as an example of a vulnerability enabling privilege escalation in LiteLLM.