The digits of pi are conjectured to be "random" in the sense of being both uniformly distributed over {0, ..., 9} and possessing no n-ary serial correlations for any n, but as far as I know, this has not been proven.

Benford's Law only applies to leading digits, and only for numbers following a certain class of distributions.

It's important to note that the "Law" is not an innate mathematical property of anything, but an observed phenomenon with mathematical underpinnings. It doesn't apply exactly to any well-defined, real-world distribution ("physical constants" is not a well-defined distribution) but it applies approximately to a good number of them.

Benford's Law is exactly true on 10^X, where X is a random variable, chosen with uniform probability, from [0, 1). It's also true with X chosen from [0, n) for any integer n, because the leading digit of 10^X relies only on the non-integral part of X. When X is chosen from some other distribution that spans many orders of magnitude (say, N(4, 1)) the distribution of the fractional part (X - floor(X)) is approximately a uniform random variable from [0, 1), so Benford's Law is approximately true.