An assumption of the linear regression model is that the response is unbounded with normally distributed errors. You might need to consider the generalized linear model instead. You can access this platform in the Fit Model launch dialog by changing the fitting personality.
Otherwise, you could save the model as a column formula, and then wrap it in the Maximum( model, 0 ) function to limit prediction to the positive number line.
You could also define a desirability function for the response with a lower bound of zero.
What is the response that you are modeling and predicting?