There is a sophisticated new flavor of accounts payable fraud taking root across corporate landscapes, and it centers entirely on the rapid monetization of artificial intelligence. Cybercriminals have uncovered a lucrative method for hijacking and monetizing stolen corporate artificial intelligence infrastructure in real time. In these schemes, threat actors sell unauthorized access to expensive large language models while the underlying companies—whose credentials have been quietly compromised—are left to absorb the staggering financial charges.
A comprehensive report released by Google’s cyber threat intelligence division exposes a glaring blind spot in modern corporate finance: artificial intelligence spending can morph into a catastrophic fraud liability long before the associated costs ever materialize on a traditional invoice. Within its findings, Google’s cybersecurity experts documented alarming incidents where fraudulent AI consumption skyrocketed past $100,000 in a single day. In some of the more severe cases, cumulative unauthorized charges at targeted organizations reached a staggering $1 million before security teams managed to revoke access.
These targeted incursions, known across the cybersecurity industry as "LLMjacking," have successfully compromised organizations spanning financial services, retail, technology, and transportation sectors across multiple continents. For corporate chief financial officers and financial controllers, these attacks expose a fundamental weakness in the existing financial controls that surround consumption-based AI infrastructure. In these modern cloud environments, an authenticated software request can automatically trigger a massive, billable expense without a single individual within the finance department ever reviewing, let alone approving, the transaction.
The result of this operational loophole is an entirely emerging category of operating expense fraud. In this space, cybercriminals successfully capture the economic value of critical corporate resources while seamlessly transferring the entirety of the financial cost directly to the victimized enterprise. It is a dynamic that heavily favors the perpetrator while imposing severe financial and operational pain on the target.
Cybercrime Discovers a Pay-as-You-Go Business Model
Traditional financial cybercrime has historically relied on direct methods such as stealing liquid capital, extracting sensitive proprietary information, or launching disruptive ransomware attacks that demand extortion payments. LLMjacking, however, introduces an entirely different vector for monetization by seizing control of infrastructure that possesses a clear, measurable, and continuous market price.
Unlike conventional ransomware campaigns, which depend entirely on a victimized organization agreeing to negotiate and pay a ransom, LLMjacking exploits operational infrastructure that automatically generates mounting charges simply by functioning. Premium artificial intelligence models typically charge for processing incoming user requests based heavily on the sheer volume of input and output tokens consumed. Armed with valid credentials, malicious actors can programmatically generate massive, relentless volumes of requests against an organization’s existing cloud accounts.
Some specialized attackers concentrate their efforts primarily on the acquisition of credentials or the deep compromise of vulnerable cloud environments. Others operate specialized illicit networks that researchers categorize as "LLM Mixers." These intermediary operations function by aggregating large pools of stolen credentials from various breaches and reselling unauthorized access to premium AI models at heavily discounted rates to downstream buyers.
One such underground service, identified as Jet Night, aggressively advertised unlimited access to advanced frontier models for a flat fee of approximately $60 per month. Google researchers successfully traced this illicit service back to an attacking entity that generated more than 500,000 application programming interface calls in less than eight hours. This single burst of unauthorized activity managed to pass nearly $1 million in token processing costs directly to the victim organization.
This operational reality is precisely what differentiates LLMjacking from conventional forms of cloud resource theft. The stolen product is not merely passive computing capacity or standard storage space. Rather, it is the victim organization’s valuable ability to purchase expensive, high-end artificial intelligence services instantaneously and on demand.
Authentication Is Becoming Crucial to Financial Authorization
According to the insights published in Google’s threat intelligence findings, malicious actors frequently bypass the need to defeat sophisticated artificial intelligence safety safeguards. Instead, they exploit familiar and persistent vulnerabilities such as exposed application programming interface keys, compromised enterprise service accounts, stolen identity tokens, and hardcoded credentials accidentally left embedded within software repositories. Once an attacker gains initial access to the environment, they can frequently expand their system privileges, activate previously dormant or unused AI services, or generate entirely new sets of administrative credentials to maintain persistence.
In one documented case of LLMjacking, attackers actively enrolled compromised enterprise accounts into automated pay-as-you-go agreements, deliberately selected premium foundation models, and initiated high-volume inference traffic. To further complicate detection and remediation, the perpetrators routinely created and deleted short-lived service accounts designed specifically to evade strict rate limits and preserve continuous access to the computational resources.
A modern cloud platform may successfully and correctly authenticate an incoming application programming interface request and subsequently bill the associated corporate account for the service rendered. However, a successful technical authentication event does not inherently establish that the business authorized the underlying financial expense. For corporate finance teams, this operational friction raises a critical question long familiar to traditional accounts payable and procurement controls: Who is officially authorized to incur a specific expense, and what technical or administrative mechanisms prevent spending from exceeding that established authorization?
Within dynamic cloud infrastructure environments, those vital financial controls increasingly need to operate directly at the granular level of service accounts, application programming interface credentials, specific model permissions, and rigid usage quotas, rather than relying on traditional purchase orders alone.
For corporate chief financial officers, the broader implication of these emerging threats is that consumption-based artificial intelligence is fundamentally altering the traditional boundary line separating cybersecurity from corporate financial control. Google’s threat intelligence analysts anticipate that the underground criminal marketplace for stolen artificial intelligence access will inevitably expand as premium models, machine learning agents, and autonomous corporate workloads see wider deployment across industries.
The exact same underlying infrastructure that empowers modern businesses to purchase intelligence instantly and scale operations seamlessly can also allow bad actors to consume those exact resources just as quickly. When the monthly cloud bill finally arrives, the cloud service provider may have delivered precisely what the compromised account requested. The lingering question for corporate leadership is whether the enterprise ever intended to buy it in the first place.
