You assume a normal curve. You define "volatility" (the market's uncertainty to the price of something) as Geometric Brownian motion with some standard-deviation / variance, and then perform differential equations on that to determine the "proper price" of an option.
After all: Options grow more valuable in times of uncertainty, and grow less valuable when everyone can accurately predict future prices.
The part that "brakes" because of negative rates is the "risk free rate", predicting the future value of money. To me, it seems like there are a variety of obvious extensions you can do to fix this problem. You could redefine Black-Scholes to operate on "Real" terms, and then convert everything back into "normal" terms later.
----------
Black Scholes doesn't "fundamentally" require postivie interest rates for the risk-free rate. It was just an assumption that they had in the original model. But as a differential equation, its pretty easy to tweak your assumptions around and fix these issues.
But any such tweak means you're no longer working with classic Textbook Black Scholes. Sure, it served as the basis of the theory, but you gotta account for more real-world variables than the original textbook model.
Interestingly, the Schrodinger equation works about the same way. Just assume a handful of reasonable and "obvious" things about physics, and out pops the equation.