Google LLMjacking Attacks Generate Over $1 Million in Fraudulent Cloud AI Charges
What Happened — Researchers from Google’s Threat Intelligence Group (GTIG) and Mandiant documented a surge of “LLMjacking” attacks in 2026. Threat actors first locate exposed LLM endpoints or steal valid API keys from poorly protected cloud assets. The stolen credentials are then sold to “LLM mixers” that relay unlimited AI requests through the victim’s accounts, inflating token consumption. In several cases, organizations were billed more than $100 k per day, with cumulative charges topping $1 M. The attackers also exploit a LiteLLM vulnerability (CVE‑2026‑40217) to run code on compromised servers and may plant back‑door accounts or SSH keys to maintain persistence.
Why It Matters for Trust & Control Assurance — The incident underscores the need for robust API‑key lifecycle management and continuous monitoring of credential use.
Who Is Affected — Cloud‑based SaaS providers, AI platform users, and any enterprise exposing LLM APIs.
Recommended Actions –
- Inventory all LLM‑related API keys and rotate them regularly.
- Enforce least‑privilege scopes and restrict network access to known endpoints.
- Deploy real‑time usage analytics to flag abnormal request volumes.
- Audit secret‑management stores for exposed credentials.
Technical Notes – Attack vectors include exposed cloud credentials, compromised service‑account keys, GitHub tokens, and the LiteLLM CVE‑2026‑40217 code‑execution flaw. The fraud is monetized via “LLM mixers” such as jet‑night.com that sell unlimited access plans. Source: BleepingComputer