That generally only happens in cases of overfitting, where the model was trained on a poorly de-duplicated data set that contains many copies of that book (or excerpts, quotes, and so forth). This is considered a flaw by AI trainers and a lot of work goes into sanitizing the training data to prevent it.
Some authors typed the first few sentences of their book and the LLM spit out the rest.
That generally only happens in cases of overfitting, where the model was trained on a poorly de-duplicated data set that contains many copies of that book (or excerpts, quotes, and so forth). This is considered a flaw by AI trainers and a lot of work goes into sanitizing the training data to prevent it.