No, it can easily happen.

- They don’t do compression by “definition”. They are designed to predict, prediction is key to information theory, so they just have similar qualities.

- Everyone wants their model to learn, not copy data, but overfitting happens sometimes and overfitting can look the same as copying.