🚀 go-pugleaf

In article <581@gec-mi-at.co.uk> adam@gec-mi-at.co.uk (Adam Quantrill) writes:
-for a less secure method, I propose to generate the random string by exclusive-
-oring any combination of n text and binary files held on computer:
-e.g. exor /bin/sh /bin/crypt /etc/termcap | exor my_text
-Does anyone have any comments as to the security of this method? It would
-obviously be more secure using a common but unique random file owned by the two
-parties, in combination with the standard system files.

Making allowances for the incorrect command in the example,
if you plan to encrypt many messages with this scheme, you
had better provide for a variable key to select the specific
XORs and starting offsets.  Otherwise one can simply "stack"
several messages and the underlying plaintext frequency
characteristics will shine through (and once the general system
is cracked, cryptanalysis of further messages is reduced to
simple decipherment).  Brute-force searches are not necessary.

Even the variable key introduces the problem of how to keep
it associated with the encrypted files (or transmitted to the
recipient, if encrypting for file transmission).  Anyone
knowing the general system who can also latch onto the
specific keys is also in the position of a simple decipherer.

Note that the XOR stream you use is no better than simply taking
yay many (truly) random bits and putting them in a protected
file, so the added complexity of combining non-random files is
rather pointless.  (Send your recipient a copy of the file on
a couriered magtape.)

I highly recommend working your way through the first two
volumes of Military Cryptanalytics (Callimahos & Friedman),
including working all the problem sets AND ESPECIALLY heeding
their advice for proposers of new cryptosystems, before trying
to design a new cryptosystem.  These books are now available
from Aegean Press (4 volumes since the appendices are bound
separately); they've done a commendable job of reprinting, and
although these books are not inexpensive they're indispensible
for the serious cryptanalytics student.  (Get the "library"
binding if you can afford it; I have the paperbacks and although
they're not bad I fear they may lose pages under heavy use,
since the pages are glued individually to the binder rather
than being sewn in signatures.)  Aegean Press also reprints
several other classics of cryptography; someone not long ago
posted an abridged list, address, etc. but I could repeat it..