Hackers are running up huge AI bills on company accounts

Criminals are breaking into company AI accounts and using them for free, leaving the victim with the bill. Google's Mandiant incident response team says the fake charges can top $100,000 a day. At some organizations, the total hit $1 million before the company cut the thieves off. The scheme is called LLMjacking.

What happened

An LLM is the kind of AI model behind chatbots. Companies pay for it by usage, counted in small units called tokens. To connect their software to these models, they use access keys, which work like passwords. If a thief gets a key, every request they make is billed to the company that owns it.

Google says the practice has grown steadily over the past two years. An August CrowdStrike report described one thief who sent about 200,000 requests in just two minutes.

How the thieves work

In one case, a group Google tracks as UNC6993 got hold of AI keys, then made more keys and turned on extra services inside the victim's cloud account. That alone ran up $100,000 in charges. In mid-2026, the same group got in through a misconfigured service account key, which is a login meant for software rather than a person.

Another thief signed victims up for pay-as-you-go plans to reach premium models. The thief then created and deleted short-lived accounts to get around usage limits, while watching the billing page to see the damage.

Some of the stolen access is resold. Google says a service called Jet Night used stolen keys to make more than 500,000 requests in under eight hours. It sells an unlimited plan for about $60 a month, while similar access would typically cost $100 to $250. Google believes the stolen keys make that price possible.

Who it affects

The risk falls mainly on organizations that use cloud AI services and access keys. Teams that leave keys in their code, or give accounts more permissions than they need, are easier targets. The material does not say how this affects people who only use a chatbot app on their own.

One trick is worth knowing for everyone. Thieves built a fake page posing as a way to join a group of AI users. It told people to paste a block of text containing their active login token, which let the thieves skip passwords and two-step login. Never paste that kind of data into a page you were sent.

What to do

The steps in the alert box above cover the main defenses: where keys are stored, who can use them, scanning code, and watching bills. John Hultquist of Google Threat Intelligence Group says locking these accounts down protects against runaway costs and also limits cheap, automated attacks on others.

What happens next

Google expects criminals to focus more on LLMjacking than on older forms of resource theft such as cryptojacking, which is hijacking computers to mine digital coins. It also sees a split in the trade. Some criminals break in, while others, called LLM Mixers, handle selling the access. Jet Night has admitted to outages, which Google thinks happen when victims spot the theft and cancel the stolen keys. The material does not say how many companies have been hit overall.