De-anonymizing Programmers via Code Stylometry

Troy Benjegerdes hozer at hozed.org
Thu Dec 31 09:41:14 PST 2015


On Thu, Dec 31, 2015 at 04:49:53AM -0800, coderman wrote:
> On 12/31/15, Georgi Guninski <guninski at guninski.com> wrote:
> > ...
> > Have they tried it on perl or assembler (or whitespace or brainfuck and
> > other esoteric languages)?
> 
> yes, even intentionally obfuscated disasm resistant mungifications of
> compiled programs.
> 
> 
> > Obfuscated perl appears hard (in case it significantly different than
> > the preimage ;) )
> 
> the key aspect is that it is well distinguished among obfuscated perl
> programs written by other coders, especially architecturally complex
> perl programs (of any variety)

Seems like if you have code that is both complex and has good performance
characteristics, then this leaves signals in the code that come from the
collective experience of the neural network(s) that created said code.

Obfuscation works by adding noise, and performance overhead.

Now the real question on my mind is how much of what I see in bitcoin's
main.cpp is obfuscation by Satoshi, or is someone going to run this 
analysis on code running at banks and figure out who he really is?



More information about the cypherpunks mailing list