Then on top of that, it hallucinates and lies. Chances for bugs or things getting snuck in increase. Since people didn’t write the code, they don’t understand it, so maintainability goes down.

This is called a "self correcting problem". If it doesn't work people will correct by either "not using it" or "learning to use it more effectively".

Do you think everyone is just going to throw their hands up and be like "well the ai done fucked up again, what're you gonna do?"