LLMjacking.

An attack in which adversaries use stolen cloud credentials to access and abuse hosted large language model services, running up large inference bills for the victim or reselling the access. It belongs to the AI & ML Security category of cybersecurity.

An attack in which adversaries use stolen cloud credentials to access and abuse hosted large language model services, running up large inference bills for the victim or reselling the access.

LLMjacking is a cloud abuse technique, named by the Sysdig Threat Research Team in 2024, in which attackers obtain stolen or leaked cloud credentials and use them to access managed large language model (LLM) services such as Amazon Bedrock or Azure OpenAI. Rather than exfiltrating data, the attacker consumes paid model inference at the victim's expense, often probing which models and quotas are enabled before driving heavy usage that can generate very large bills. Stolen keys are frequently funneled through reverse-proxy tooling so the operators or their customers can query premium models anonymously, effectively reselling someone else's cloud account. LLMjacking is conceptually similar to cryptojacking but targets AI inference capacity instead of cryptocurrency mining, and it is mitigated by guarding credentials, restricting model permissions, and monitoring for anomalous usage and cost spikes.

Defences for LLMjacking typically combine technical controls and operational practices, as detailed in the full definition above.

Common alternative names include: LLM jacking, LLM hijacking.