I don't see a practical way to prove this though. The software is developed behind closed doors and unless someone explicitly leaks this information, it's just

I don't see a practical way to prove this though. The software is developed behind closed doors and unless someone explicitly leaks this information, it's just a caption feature working really well. Amazon could claim they trained it on a legit data set they had all the necessary rights for. Or am I missing something and they're openly admitting they used it?