What is LLMjacking
Google and Mandiant researchers have documented a commodity underground market for stolen AI platform credentials. Attackers compromise cloud environments, extract API keys for services like OpenAI, Anthropic, and AWS Bedrock, and then either use those keys directly for large-scale inference tasks or resell them in private Telegram channels. One leaked key can generate thousands of dollars in fraudulent inference costs within hours.
Why AI API keys are your new cloud root credentials
Traditional cloud credential theft targeted compute and storage. LLMjacking adds AI inference to that target list. The danger is not just the direct cost of fraudulent usage. An attacker with your AI API key can also observe the prompts your application sends to the model, extract business logic embedded in system prompts, and potentially influence model outputs if the key grants access to fine-tuned models or assistants. Treat AI API keys with the same access controls you apply to production database credentials.
How the compromise happens
Most LLMjacking incidents begin with exposed environment files, leaked secrets in public repositories, or compromised CI/CD pipelines. Attackers scan continuously for exposed .env files and misconfigured object storage buckets. Once a key is found, automated tooling validates it within seconds and queues it for exploitation. The window between exposure and exploitation is measured in minutes.
What to do now
Audit every AI API key in your environment. Rotate any key that has been in use for more than 90 days or that was present in a repository at any point. Enable spending limits and anomaly alerts on your AI platform accounts. Treat AI API keys as secrets in your secrets manager, not as configuration values in environment files. If you use Infisical, HashiCorp Vault, or AWS Secrets Manager, migrate your AI keys there immediately.
The durable lesson
Every new capability platform your organization adopts creates a new credential surface. AI platforms arrived fast and the security hygiene for managing AI credentials has not kept pace with the security hygiene for managing cloud credentials. The fix is conceptual as much as technical: AI assets are infrastructure, and AI API keys are infrastructure credentials. Treat them accordingly.
Gigia Tsiklauri is a Security Architect and founder of Infosec.ge. Get in touch if you want to discuss AI security and credential hygiene with her.