> we do not build or train these models in such a way that they could learn, memorise, or be able to reproduce some part of Customer Data They don't "build" th

> we do not build or train these models in such a way that they could learn, memorise, or be able to reproduce some part of Customer Data They don't "build" them this way (whatever that means) but if training data is somehow leaked, they're off the hook because they didn't build it that way?