LLMJacking: Stolen AI Access Is Costing Victims Up to $100,000 a Day

Criminals are stealing access to company AI tools and leaving the victims with the bill. According to a Monday report from Google's Mandiant incident responders, as covered by The Register, these fraudulent token charges can exceed $100,000 a day. At some organisations the total reached $1 million before access was revoked. The tactic is called LLMJacking, and it has been growing steadily over the past two years.

The report describes several cases. A group Google tracks as UNC6993 gained access to LLM API keys, then created more keys and enabled extra services inside the victim's cloud projects, running up $100,000 in charges. Another thief signed victim accounts up to Pay-As-You-Go agreements to reach premium AI models. That attacker used a "burn and rotate strategy", creating and deleting short-lived service accounts to get around rate limits while watching the billing page to track the financial damage. In a third incident, an intruder made more than 500,000 LLM API calls in under eight hours using stolen third-party keys to power a token routing service called Jet Night.

Google's John Hultquist said attackers want to use this technology to gain a scale advantage over defenders, and that locking down these resources protects organisations from runaway costs and helps limit cheap, autonomous attacks. Google's analysts believe Jet Night relies on stolen keys to make a profit, since it offers an unlimited plan for about $60 USD a month, well below the typical $100 to $250 market price.