Google Reports LLMjacking Attacks Generating Over $1 Million in Fraudulent Cloud AI Charges

Google has reported that LLMjacking attacks have collectively generated more than $1 million in fraudulent cloud charges, according to BleepingComputer. In LLMjacking, threat actors hijack access to large language model services running in a victim's cloud environment. Some affected organizations reportedly lost more than $100,000 in a single day.

Why It Matters

LLMjacking commonly relies on stolen or exposed cloud credentials and API keys. Attackers then use the victim's cloud AI resources to run their own workloads. Cloud LLM billing is usage-based and can scale quickly, so costs can mount fast and may go unnoticed until an invoice arrives. Organizations running generative AI workloads in the cloud are potential targets, especially those with broadly scoped or poorly rotated credentials.

What Defenders Should Watch For

• Audit cloud IAM roles and API keys tied to AI and LLM services for broad permissions, unused keys, and inconsistent rotation.
• Monitor cloud billing and usage dashboards for unusual spikes in LLM API calls, including from unfamiliar regions, accounts, or at odd hours.
• Hunt for credential exposure such as leaked keys in public repositories, CI/CD logs, container images, or misconfigured secrets managers.
• Set billing alerts and usage limits on LLM services for cost control and early warning.
• Review cloud audit logs for unfamiliar principals or service accounts calling generative AI APIs.

This is a developing story based on reporting from Google. Further technical details may emerge. For the original report, see BleepingComputer's coverage.