Yet, the models are objects built by humans. Any and all responsibility inevitably falls on the humans that built something with those flaws, and/or allowed them to exercise said flaws.
There’s no other explanation. Complexity of the algorithm or the application doesn’t change the reality.