Google: LLMjacking attacks rack up $1 million in cloud AI charges
Our summary
Google's Threat Intelligence Group and Mandiant report that so-called LLMjacking schemes are generating massive fraudulent cloud bills for victim organizations, with some losing over $100,000 in a single day and totals reaching as much as $1 million. Two cooperating criminal clusters are involved: one that scans for exposed LLM endpoints and another that harvests valid API keys, service account credentials, GitHub tokens and session tokens from poorly secured systems. The stolen access is then resold through so-called "LLM mixer" services, such as Jet Night and newapi[.]pro, which offer cut-rate subscriptions while quietly routing customer prompts through hijacked accounts — in one case generating over 500,000 API calls in under eight hours and nearly $1 million in charges for the victim. Google also flagged a cluster tracked as UNC6993 aggregating stolen keys for resale, and noted active exploitation of CVE-2026-40217, a remote code execution flaw in LiteLLM's custom-code guardrail testing endpoint, urging organizations to tighten IAM controls, rotate secrets, enforce phishing-resistant MFA, and monitor for shadow AI usage and spending anomalies.
BleepingComputer publishes this story on its own site; we link to it rather than reprint it.
Worried this affects your company?
Discuss a security assessment of your internet-facing systems. Scope agreed before testing.