LLMjacking turns stolen AI access into million-dollar cloud bills

The rise of LLMjacking

A new type of fraud is emerging that targets the infrastructure behind artificial intelligence. Cybercriminals are now finding ways to monetize stolen corporate AI access in real time, selling unauthorized use of large language models (LLMs) to third parties. While these criminals profit from the sales, the original companies whose credentials were stolen are forced to pay the resulting cloud bills.

This specific threat is being called LLMjacking. It differs from traditional cybercrime because it does not necessarily involve stealing money directly or demanding a ransom payment. Instead, it exploits the pay-as-you-go business model of cloud computing, where services are billed automatically based on the volume of data processed.

According to a report from Google’s cyber threat intelligence division, this is becoming a significant blind spot for corporate finance. The report highlights that AI spending can become a major fraud liability long before a company even sees an invoice for the activity.

How the fraud works

The mechanics of LLMjacking rely on exploiting existing weaknesses in how companies manage their cloud environments. Attackers often find exposed API keys, stolen identity tokens, or credentials that were accidentally left in software repositories. Once they gain entry, they can activate unused AI services or create new credentials to maintain their presence.

In some instances, criminals have enrolled compromised accounts into pay-as-you-go agreements to access premium foundation models. They then generate high-volume traffic to create massive amounts of inference, which is the process of the model generating a response. To avoid being caught by rate limits, some attackers even create and delete short-lived service accounts to keep the traffic flowing.

One specific example involved a service called Jet Night, which advertised unlimited access to frontier models for about $60 per month. Google researchers linked this service to an attacker who generated more than 500,000 API calls in less than eight hours. This activity passed nearly $1 million in token costs directly to a victim organization.

Impact on corporate finance

For Chief Financial Officers (CFOs), LLMjacking represents a breakdown in traditional accounts payable controls. In many cloud environments, a software request can trigger a billable expense without any human in the finance department ever approving the transaction. This creates a situation where the company is paying for a service it never intended to purchase.

The problem is that cloud platforms are designed to fulfill requests as long as the authentication is valid. If a criminal has the right keys, the platform correctly bills the account for the work performed. As a result, the financial control must move beyond simple purchase orders to include monitoring at the level of service accounts and model permissions.

The scale of the problem is already visible across various industries. Google documented incidents where fraudulent AI consumption exceeded $100,000 in a single day. These attacks have already impacted companies in the transportation, retail, technology, and financial services sectors across multiple continents.

The expanding threat landscape

The marketplace for stolen AI access is expected to grow as businesses deploy more autonomous workloads and premium models. Because the infrastructure allows for instant purchasing of intelligence, it also allows criminals to consume that intelligence just as quickly. This creates a high-speed environment where the cost of a breach can escalate in a matter of hours.

Security experts suggest that the boundary between cybersecurity and financial control is blurring. Companies need to treat AI usage quotas and API credentials as critical financial assets. Without these specific controls, organizations remain vulnerable to attackers who can drain resources by simply running high-volume requests.

As more firms adopt AI to improve efficiency, the incentive for criminals to target these systems increases. They are moving away from simple data theft toward capturing the economic value of a company's cloud resources. For businesses, the challenge is ensuring that every dollar spent on AI is an intentional business decision.