A language model does not look things up. It writes what sounds likely.
Ask ChatGPT for case law and you get a convincing answer: a court, a date, an ECLI in the right format. But the model never read that ruling. It predicts characters that look like an ECLI. Sometimes the ruling exists, sometimes it is about something else, sometimes it does not exist at all.
The same goes for statutes. The model knows a version from its training data, not the text in force today, and not the text in force on the date of your facts. A quotation in inverted commas is often a paraphrase.
Whoever puts that in a court document answers for it. In the American case Mata v. Avianca (2023), lawyers filed a brief citing six rulings that ChatGPT had invented. The court fined them 5,000 dollars.