LLM-jacking is the abuse or resale of legitimate access to large language model services. Instead of breaking the model, an attacker steals valid credentials, API keys, or session access and uses them as if they were the real owner. Because the access is genuine, the traffic may look normal and can be hard to distinguish from approved use.
This matters because AI access is expensive and often tied to billing, quotas, and sensitive workflows. An attacker can run prompts, automate abuse, or resell the access while the victim pays the bill and absorbs the operational risk. Defenses focus on strong secrets management, least-privilege access, key rotation, session invalidation, and monitoring for unusual usage, integration activity, or billing spikes.