When should you not use logistic regression GLM with binary response variable?

Question

I came across a set of data that does not seem to "flip" from 0 to 1 (binary response variable) when the predictor variable increase. Thus, this is the plot I got