I think there are two distinct types of "bad" here, and to me, one is way more interesting than the other.
If we mean, that this is dangerous: Yes, of course. It's an obvious, somewhat dangerous error (and I say "somewhat", only because I assume currently everyone participating is cognizant of the fact, that this is in some kind of testing stage)
But I think the more interesting question is: How quickly can this issue (and others like it) be fixed? If the answer at this point is "we will have to retrain the entire model and just hope for the best" that sounds, like, really bad.
If we mean, that this is dangerous: Yes, of course. It's an obvious, somewhat dangerous error (and I say "somewhat", only because I assume currently everyone participating is cognizant of the fact, that this is in some kind of testing stage)
But I think the more interesting question is: How quickly can this issue (and others like it) be fixed? If the answer at this point is "we will have to retrain the entire model and just hope for the best" that sounds, like, really bad.