[118069] in Cypherpunks
Re: The "ECHELON" Chip
daemon@ATHENA.MIT.EDU (Martin Minow)
Sat Sep 18 23:12:51 1999
Mime-Version: 1.0
Message-Id: <v04210101b409fc0605ca@[63.193.122.223]>
In-Reply-To: <19990918212028.52899.qmail@hotmail.com>
Date: Sat, 18 Sep 1999 19:51:55 -0700
To: "Gary Jeffers" <jeffersgary@hotmail.com>
From: Martin Minow <minow@pobox.com>
Cc: cypherpunks@cyberpass.net, cryptography@c2.net
Content-Type: text/plain; charset="us-ascii" ; format="flowed"
Reply-To: Martin Minow <minow@pobox.com>
In a note to Cypherpunks, "Gary Jeffers" <jeffersgary@hotmail.com>
offers some suggestions for adding a back door to future cpu chips
that, to simplify his note, would bypass existing memory protection
and similar schemes.
Here's an even simpler cpu hack:
1. A manufacturer adds a cryptographically strong random number
generator to its cpu architecture that is judged by the
cryptographic community as providing truly random numbers.
2. The only public api (on or off the chip) passes the raw
data stream through a SHA1 whitener for added security.
(I'm not a cryptographer, but don't understand why, if
the input is truly random, the output needs additional
whitening, but it's important because of step 3 below).
3. The manufacturer also adds an unpublished instruction
orinstruction sequence that changes the random number
generator from "true randomness" (whatever that is) to
X = (X * 5) + 233 seeded with the current time of day and
re-seeded every second. Since the only user-visible output
has been SHA1 whitened, this substitution will be extremely
difficult to detect (Especially since the initial vetting
of the random number generator will be done with cpu's that
were not hacked.
4. The appropriate Three Letter Agency slips the trojan horse
instruction into the target's computer and can read encrypted
files as long as they have a reasonable estimate of the
time they were encrypted.
The above seems too simple to be realistic, but does explain
why SHA1 was needed to "whiten" a truly random sequence.
Martin Minow
minow@pobox.com