Hackers have learned to make money from corporate AI assistants. Attackers hack into corporate accounts, gain access to expensive AI models, and resell them to other users at low prices. Companies that don't even suspect the "uninvited guests" then receive bills for hundreds of thousands of dollars.
Hackers have learned to make money from corporate AI assistants. Attackers hack into corporate accounts, gain access to expensive AI models, and resell them to other users at low prices. Companies that don't even suspect the "uninvited guests" then receive bills for hundreds of thousands of dollars.
Researchers from Google Threat Intelligence Group (GTIG) and Mandiant reported on this scheme.
This type of fraud is called LLMJacking. Attackers steal special access keys to AI services or find them where developers have accidentally left them open. For example, in code on GitHub. Sometimes it is also enough to gain access to a corporate cloud account.
After that, hackers can use expensive neural networks, send them thousands of requests, and even run powerful servers. And all this at the expense of the company that owns the hacked account.
Hackers are also pooling access to neural networks from different compromised accounts and then offering it to anyone at significantly lower prices. Google researchers have found a whole "black market" for such services.
For example, Jet Night promises supposedly unlimited access to advanced AI models for about $60 per month. According to researchers, this level of access would normally cost $100–250. Google suspects that Jet Night can afford such prices precisely because of stolen keys.
In one case under investigation, an attacker sent over 500,000 queries to neural networks in less than eight hours, resulting in the victim company being charged nearly $1 million for using AI.
Additionally, companies' bills can add up very quickly. Researchers have also documented cases where hackers used more than $100,000 worth of other people's AI resources in just 24 hours.
Google believes that such a business could become even more attractive to cybercriminals than hidden cryptocurrency mining. After all, stolen access to powerful neural networks can be quickly resold, and all costs can be passed on to the owners of the hacked accounts.
To avoid paying for other people's conversations with AI, companies are advised to monitor their access keys, check for unusual activity, and set spending limits.
Previously, dev.ua wrote about a Mexican startup that received a bill for over $82,000 in just two days due to a stolen Gemini API key.
