What is LLMjacking?

LLMjacking is when attackers use stolen cloud or AI keys to run AI models on someone else's account, leaving the victim with the bill.

Share

LLMjacking is when attackers use stolen cloud or AI keys to run AI models on someone else's account, leaving the victim with the bill.

Why it matters

AI agents need keys to call models, and those keys often sit in hosting platforms, code and developer laptops. A stolen key can be switched on for expensive models, resold to others, or used to run the thief's own agents, and the charges land on the key's owner.

Example

On 4 November 2026, Google's Threat Intelligence Group described attackers using stolen cloud credentials to enable AI model services and resell the access, running up bills of up to $1 million.

How to defend against it

  • Keep AI model services switched off in cloud accounts that don't need them.
  • Use short-lived, per-workload credentials instead of long-lived keys, and cap spending per key.
  • Alert when model access is enabled or model logging is deleted by anyone other than your infrastructure pipeline.
  • Rotate keys stored in any hosting platform that reports a breach. See the AgentSec Brief checklist.

Environment-variable secrets, denial of wallet, token and privilege compromise.

Sources

From AgentSec Brief, the daily brief on securing agentic AI. Subscribe free at https://www.agentsecbrief.com