A student model is a smaller AI model trained to imitate the behavior of a larger teacher model. Instead of learning only from labeled ground truth, it learns from the teacher’s outputs, which can include probabilities, explanations, rankings, or generated text. This is common in model distillation, a compression method that reduces compute cost and makes advanced capabilities easier to deploy on limited hardware.
In cyber security, student models matter because the outputs of a powerful system can become reusable intelligence. An attacker may query a teacher model at scale, collect responses, and train a student model that copies useful behavior without direct access to the original weights or code. Defenders watch for repeated prompts, bulk harvesting, and unusual automation because these patterns can indicate model extraction or capability transfer. Organizations that expose valuable models should protect prompts and completions, enforce rate limits, and monitor for suspicious reuse of output data.