mm... not so easy in my pov...
DeepL is not public entity neither, and the database build and used to train the "brain" is not open to scrutiny nor correction (appart small use changes)
with this (type of) service becoming used, what about f.ex. gender bias in the database, while society would like to evolve and change way to speak in french (f.ex.) ...
this kind of services should become commons, like the speach recognition engine that @Mozilla has released.
yes, you better do, their data policy while not super evil, is not really helping neither...https://www.deepl.com/privacy.html
at least it's quite clear "you work for them" by submitting your data and corrections, so they train the brain in return (maybe some way to hack it with adversary training :p )