I used to build automated voice systems and chatbots for a large retail organisation, this was in the days of traditional NLU but before LLM stuff.
I mandated every bot and automated journey we deployed had to have a number of things in place to prevent this problem--really trivial stuff like "if there 2 or more failure states in succession (error or customer says something like 'no not that', offer a human agent.", or if a human agent is requested but no automation was attempted, encourage the customer to try automation first but don't force them to stay in automation.
There was always pressure from other parts of the org (marketing, financial types etc.) to contain people inside bots in a hostile way all to "reduce costs" which is a false economy, because customers either leave out of frustration or these tasks generate full blown complaints which ended up costing more to manage.
It's tough to push back on that stuff for some dev teams, I'm guessing they get outranked a lot. I was lucky to have board-level backing of my approach