That's absolutely usability. To show what will happen if something like "data" on a service like ChatGPT is touched, to prevent outcomes the user did not want, to prevent accidental data loss through guidance and safeguards. It's usually not sloppyness when users run into those situation, but bad program design. Maybe also here: ChatGPT now does not remove the old interactions when the data consent setting is changed, and shows prominent warnings before the removal of interaction threads, which is a separate option, according to a german article about this (https://www.notebookcheck.com/ChatGPT-Professor-verliert-zwe...). It is likely this was changed in the meantime, to better the usability.

Or the user really was surprised that deleting all interactions meant deleting all interactions. Then your position is a bit more understandable, but even then - mistakes happen and an undo would still be good.