I would complain to your foundry modeling department.  It could mean your whole simulation is corrupted.

It could be that the subckt has components for contact/end resistors and a body resistor, and only the end resistors are negative; if the body resistor is large and the end resistors aren't very negative, it could be OK.

The quick way to fix it is to replace the subckt model with a basic linear resistor -- assuming the nonlinearity isn't very strong and/or the resistor isn't in a performance-critical circuit.

Another possibility would be to change the geometry of the resistor such that the nonlinearity is reduced -- ie, double the W and L.  Unfortunately, you might end up with a resistance that is still wrong, maybe it's 1/100 of the nominal (unbiased) resistance, but because it's not negative, you don't get any warnings.