NEWS

Google Warns LLMjacking Is Costing Enterprises Up to $1 Million in Cloud Charges

Google GTIG and Mandiant say hijacked cloud and LLM credentials are resold via 'LLM mixers', with fraudulent bills topping $100,000 a day.

Dylan H.

News Desk

November 3, 2026
2 min read
Google Warns LLMjacking Is Costing Enterprises Up to $1 Million in Cloud Charges

Overview

Google Threat Intelligence Group (GTIG) and Mandiant warned on Monday that LLMjacking, the hijacking of enterprise access to large language models, has become a profitable criminal business. Individual incidents have produced fraudulent charges above $100,000 in a single day, and cumulative bills at some organizations reached $1 million before access was revoked.

The $1 million figure comes from Google's statement. No victim invoice has been published.

How the Attacks Work

LLMjacking is resource hijacking adapted for the AI era. Unlike cryptojacking, it can generate large spikes in charges within hours because frontier model inference is expensive.

Reported entry points:

  • Exposed or compromised cloud credentials and service account keys
  • Stolen GitHub tokens and active session tokens
  • Leaked LLM API keys
  • CVE-2026-40217, a LiteLLM flaw allowing arbitrary code execution through its guardrail testing endpoint

The "LLM Mixer" Resale Economy

Stolen access is sold to "LLM mixer" services such as Jet Night, which charge as little as $60 a month for unlimited access. They route customer requests through victims' hijacked accounts. One Jet Night operation made more than 500,000 LLM API calls in under eight hours, passing nearly $1 million in token costs to the victim.

Google also tracks UNC6993, which created additional API keys inside victim cloud projects and ran up over $100,000 in charges. In some cases attackers hosted unauthorized proxy endpoints and agentic workloads on the victim's infrastructure.

Mandiant responded to multiple incidents this year in financial services, retail, technology, and transportation.

Recommendations

  1. Set hard budget caps and quotas on every AI service, with alerts on spend velocity, not only monthly totals.
  2. Inventory and rotate LLM API keys and cloud service account keys. Prefer short-lived, workload-bound credentials.
  3. Scan repositories and CI logs for leaked keys and tokens.
  4. Patch LiteLLM and restrict access to guardrail and admin endpoints.
  5. Monitor for new API key creation and unexpected model invocations in cloud audit logs.
  6. Alert on unusual egress and new proxy endpoints in AI-enabled cloud projects.

Sources