Excellent answers above (by CactusCake and TTT, respectively) that explain the two key points that:
1) The interest rate for a group is based on probability of default. It's a group characteristic, calculated based on historical data.
2) The interest rate doesn't increase the probability of default, because if they give you a higher interest rate they also approve a lower loan amount. Your burden is the same whichever interest rate you get and doesn't affect your ability to repay.
I would just like to tie them in.
The key thing to understand is the difference between the individual and the group. It's really a bit as if the single customer is invisible to the bank. This is counterintuitive, because he/she provides so much data upon application. Yet despite all that, the bank cannot figure out whether he's going to default or not. The only thing the bank can do is try to match the individual to a segment of existing customers with known historical default rates. So the bank will say (simplified): "OK, this guy is 25 years old, so he's in our 20-30 year-olds cohort, which has a default rate of 10%. For that group to be profitable to us, we have to charge them 8% each." (See CactusCake reply for a simple calculation.)
People think, mistakenly, that when a bank examines their application it is focusing on them and trying to figure out whether they will default. It's not really true. Imagine a bank with zero customers. No matter how many data points about a person it had, and how hard it looked, it simply could not arrive at something like "this guy is 35% likely to default". Only after having historical data can a bank try to match the new customer to an existing group and derive the probability of default from that.
So the sequence is:
Look at customer --> figure out which group he belongs to --> check historical probability of default of that group --> set the interest rate --> adjust and approve the loan amount
It is not:
Look at customer --> reject or approve the loan amount --> set the interest rate (and thus increase/decrease probability of default)