There are a ton of methods to probe training data from a trained model, both for open and closed-source models.