Attackers steal the keys that business applications use to access AI and resell that access. Google has seen victims lose more than $100,000 in a single day.
The Google Threat Intelligence (GTIG) and Mandiant teams describe a phenomenon called “LLMjacking”: someone obtains the key a company uses to pay for an AI service (OpenAI, Gemini, Claude, or the cloud platform hosting them) and uses it at the company’s expense. A single attacker made over 500,000 requests in less than eight hours; at some organizations, cumulative losses reached one million dollars.
Keys end up in the hands of attackers in two ways: either they are stolen from unprotected computers and servers (from scripts, configuration files, code repositories, or compromised sessions), or the company’s AI application is left open on the internet. Access is then resold cheaply as an “unlimited” subscription—and the bill goes to the key’s owner.
Why this matters to you: A small business doesn’t need a large-scale project to have a tool like this. A chatbot on the website, an automation tool that summarizes emails, or a script written by a team member is enough. Usually, no one remembers where they left the key, and the account has no spending limit.
What you need to do:
- Log in to the company’s paid AI service account and set a monthly spending limit, plus an email alert at a low threshold; without a limit, a stolen key could result in an unlimited bill.
- Ask the person who developed the automation or chatbot for a list of the keys used: a separate key for each application, stored outside of scripts and emails, changed when an employee leaves, and deleted when the application is no longer in use.
A spending limit is the most cost-effective form of protection: it doesn't prevent theft, but it turns a loss of thousands of euros into an alert costing just a few euros.
Sources:
This article was generated with AI assistance.