Cybercriminal groups have adopted a new tactic called LLMjacking, exploiting stolen or exposed AI API credentials to run unauthorized large language model queries, leading to skyrocketing cloud expenses that can reach six figures in minutes.
- Attackers hijack AI API keys to run costly LLM queries on victim clouds
- Bills can exceed $100,000 within minutes due to high-volume model usage
- Specialized criminal groups resell and route stolen AI access for profit
What happened
Threat actors have turned their attention to exploiting large language model services by stealing API keys and cloud identity tokens such as IAM credentials and GitHub access tokens. By gaining entry to company cloud environments through these compromised tokens, attackers are able to escalate privileges and maintain persistent access to target infrastructure.
Once inside, attackers conduct LLMjacking—the unauthorized use and monetization of AI foundation models through techniques like token proxying and automated model streaming. This abuse of AI subscriptions has allowed criminals to accumulate cloud charges in excess of $100,000 in just a few minutes by running continuous, high-volume inference workloads.
Why it matters
LLMjacking represents a significant financial risk to organizations utilizing AI cloud platforms, as malicious actors can anonymously run large-scale AI operations at the expense of the victim. Unlike the slower drain seen in cryptojacking, LLM abuse can rapidly generate massive bills, causing severe accounting and operational headaches.
Additionally, this activity signals a maturing cybercrime ecosystem focused on AI abuse, where roles have specialized—some groups focus on initial compromise, while 'LLM Mixers' manage monetization by proxying and reselling unauthorized AI access. This evolves the threat landscape, highlighting the need for tighter cloud security and AI API key management.
What to watch next
Security teams should monitor for signs of unauthorized AI API usage and strengthen cloud access controls to prevent token theft or exposure. Implementing strict token lifecycles, usage auditing, and anomaly detection around AI service calls will be key defenses against LLMjacking schemes.
Meanwhile, as AI adoption grows across industries, expect threat actors to increasingly prioritize hijacking these expensive resources. Observing developments in both attack methods and defensive technologies will be crucial for organizations to safeguard their AI environments and manage cloud costs effectively.