You can normalize data in order to isolate the effects of a given variable. Essentially, you just change the question from 'Are android users more likely to default on a loan?' to 'Given that all other factors are roughly equal, are android users significantly more likely to default on a loan' and then you do some complex math I don't understand to determine whether income disparities are actually a factor or if the model phone you own affects your likelihood of repaying a loan without regard to your income.

Think about it this way - banks are prohibited from asking certain questions. However, if they have a strong correlation with another factor, then they can ask that and get the same result.