I think this article is missing a major point, or perhaps should be titled "Some Non-Scary Machine Code Isn't Scary". It argues that machine code isn't scary, by building a one-to-one mapping from machine code to assembly code, and then taking it as given that assembly code isn't scary. But it uses two examples -- 32-bit ARM and x86-64 -- where this one-to-one mapping isn't valid. When in Thumb mode for (some flavors of) ARM, even when you know you're in thumb mode, instructions can be a mix of 16 and 32 bits. And in x86 world, of course, instructions can be a wide range of widths. What that means is that if you're given a chunk of memory that is known to contain executable instructions... you /can't/ build a one-to-one mapping to assembly without knowing where all of the entry points are. For well-formed code you can often exclude almost all possible entry points as invalid, and maybe even end up with only a single one... but it's perfectly possible (and quite fun) to write machine code that has valid, different behavior for different entry points to the same byte sequence. There's no way to reduce this type of machine code to meaningful assembly, and it should be considered scary.