Hackers Are Turning Stolen AI Access Into a Fast-Growing Black Market
Cybercriminals are increasingly targeting corporate AI accounts and cloud credentials to run costly artificial intelligence workloads without paying for them. The tactic, often called LLMjacking, is becoming a major cybersecurity threat as businesses rely more heavily on large language models, cloud AI platforms, and GPU-powered services.
According to Google’s Threat Intelligence Group, attackers can generate massive cloud bills in a very short time. In some cases, stolen AI access has reportedly led to charges exceeding $100,000 within minutes. Unlike traditional cryptojacking, where attackers secretly mine cryptocurrency and slowly increase a victim’s expenses over time, LLMjacking can create immediate and severe financial damage.
How LLMjacking works
LLMjacking usually begins when attackers find exposed AI endpoints or steal credentials that grant access to cloud environments. These credentials may include API keys, identity tokens, service account keys, GitHub personal access tokens, or active session tokens.
Once cybercriminals gain access, they often try to expand their control inside the victim’s cloud infrastructure. They may search for AI services, increase privileges, create new accounts, or enable additional cloud features. Their goal is to gain stable, long-term access to expensive AI resources while avoiding detection.
After that, the stolen access can be used in several ways. Some attackers run their own AI workloads through the victim’s account. Others route AI requests through compromised systems or sell unauthorized access to buyers on underground markets. This allows customers on the black market to use premium AI models and cloud computing resources while the victim pays the bill.
Why AI accounts are becoming a valuable target
The rapid growth of generative AI has made access to powerful language models and GPU infrastructure highly valuable. Premium AI models, large-scale inference tasks, and cloud-based machine learning services can be expensive, especially when used at high volume.
This creates an opportunity for cybercriminals. Instead of stealing data alone, attackers can now steal access to AI infrastructure and turn it into a paid service. Google describes this as a growing form of resource hijacking, where criminals monetize foundation models through proxying, automated model usage, and GPU provisioning.
In one reported case, a threat group obtained LLM API keys and activated services inside a victim’s cloud project, quickly producing more than $100,000 in charges. In another incident, attackers enrolled compromised accounts in pay-as-you-go plans, selected premium AI models, and launched high-volume inference operations.
To bypass limits and remain active longer, some attackers created and deleted temporary service accounts repeatedly. They also monitored the victim’s billing dashboard to track usage and adjust their activity.
A new cybercrime ecosystem is forming
Google warns that LLMjacking is no longer just a simple abuse technique. It is developing into a specialized cybercrime market.
Some groups focus on breaking into corporate systems and stealing credentials. Others operate as so-called “LLM Mixers,” which act as gateways that combine stolen credentials, distribute AI requests, and resell access to underground customers. This makes it easier for buyers to use stolen AI services without directly compromising a company themselves.
As demand for artificial intelligence continues to rise, stolen AI access is likely to become even more attractive to attackers. Businesses using cloud-based AI tools, large language models, and GPU resources may face growing risks if their credentials are not properly secured.
How organizations can reduce the risk
Companies should treat unauthorized changes to AI accounts, cloud services, and billing settings as urgent security events. Strong visibility into AI usage is essential, especially for organizations running large-scale AI workloads.
Security teams should closely monitor API activity, unusual spikes in model requests, unexpected service activations, and abnormal billing increases. Sudden changes in AI usage may be an early sign that attackers are abusing cloud resources.
Google also recommends following the principle of least privilege. This means users, service accounts, and applications should only have the permissions they truly need. Overly broad access can make it easier for attackers to escalate privileges and take control of valuable AI resources.
Other important protections include rotating exposed API keys, securing cloud credentials, monitoring identity tokens, reviewing access logs, and limiting who can enable expensive AI services. Organizations should also pay close attention to exposed endpoints and leaked secrets in code repositories.
The rise of LLMjacking shows that AI infrastructure is now a prime target for cybercriminals. As companies continue adopting generative AI, protecting cloud credentials and AI accounts is becoming just as important as securing customer data, internal systems, and financial assets.
