I think it's because they've been trained by copying answers from ChatGPT. They're not really very copyrighted after all.

Though the other day I saw someone demonstrate this with Google's Gemini through the API, so maybe it is just picking up conversation traces off the internet.