Because it's not completely true. If a CPU is microcoded, then it's accurate to say "assembly is interpreted" because every instruction is effectively an address into a lookup table of microinstructions. But in a non-microcoded (e.g. purely RISC) CPU, the bits of the instruction word are effectively enable and data lines to a bunch of hardware logic gates and flip-flops, which cause register transfers and arithmetic operations to happen. In this case, the ones and zeros in the instruction word are voltage levels on logic gates. Calling the latter "interpretation" is a stretch.

To be fair, there aren't many pure RISC implementations around these days. Most everything has some degree of microcode involved, so to that extent you're right.