Inference

Inference is the runtime act of using a trained model: your prompt goes in, computation happens on the provider's (or their cloud partner's) hardware, and the response comes back. It is distinct from training — inference does not itself change the model.

The distinction anchors most AI privacy claims. "We don't train on your data" limits one use; it says nothing about how long inference inputs are retained, who can review them, or where the processing happens. Those are separate questions with separate answers per tier.