To be fair, it may yet turn out that MD5 is (and has always been, we just didn't know it yet) a low-quality hash that can be trivially preimage-attacked. I think that's unlikely[0], but I can easily imagine someone reasonable being a bit more pessimistic. I certainly wouldn't willingly rely on MD5 preimage resistance for anything important.

0: I'd be willing to bet money at 1:1 odds that MD5 preimages still cost at least 2^96 bit operations (out of a nominal 2^127 hash invocations) 5, 20, or 100 years from now.