Criminals are running up AI bills on stolen company logins, some topping US$100,000 in a day, like a stranger filling up on your fuel card
By Brett Gavaghan, Founder of Adaptd ·
Google's Mandiant and Threat Intelligence Group say attackers are breaking into company cloud accounts, switching on AI models and leaving the owner with the bill.
The short version
- Attackers are stealing cloud and AI logins, switching on AI models and leaving the account owner with the bill.
- Mandiant and Google Threat Intelligence saw cases over US$100,000 in a day, including in Asia Pacific.
- Turn on spending caps, multifactor sign-in and key limits for any AI or cloud account your business uses.
Most of us worry about AI giving a wrong answer. Fewer of us worry about someone else using our AI account all night and sending us the bill.
What is happening with stolen AI access?
Criminals are breaking into company cloud accounts and using them to run AI models at the owner's expense, according to a report from Mandiant and the Google Threat Intelligence Group covered by IT Brief on 3 November 2026. They get in with stolen passwords, hijacked sessions or access keys left exposed, including keys found in public GitHub code.
In several cases the victim had never switched those AI services on. The attackers did it for them, then used the access themselves or sold it on through cut-price resale services the researchers call "LLM Mixers".
How much damage can it do?
Fraudulent charges topped US$100,000 in a single day in some cases. One reseller pushed almost US$1 million in model costs onto a victim while charging its own customers about US$60 a month for unlimited access. In another case, more than 500,000 AI requests ran in under eight hours.
Victims were in North America, South America and Asia Pacific. Unlike older cloud abuse, which builds slowly, this kind of bill can land overnight.
What should your business do about it?
Start with any business account that can run AI on a usage bill: a cloud account, an OpenAI or Claude developer account, or an automation tool holding an API key. Put a monthly spending cap and a billing alert on each one, and turn on phishing-resistant multifactor sign-in.
My advice: if a contractor or developer ever set up an API key for you, ask where it lives and what it can do. Limit each key to the models and usage it needs, and delete the ones nobody can explain. If you only use ChatGPT or Claude through a normal monthly plan, your risk is lower, but the same strong sign-in rules apply.
A spending cap won't stop a break-in. It does stop the break-in costing more than the rest of the year's IT budget.
Sources
Wondering what this means for your business?
That's exactly what the discovery call is for.