-----BEGIN PGP SIGNED MESSAGE-----
Proposition 1:
All remailing schemes are vulnerable in the case that all remailing
sites in the chain are compromised before transmission.
Proposition 2:
With ARAs and direct transmission (one recipient at each hop), if the
first n-1 hosts in a chain are compromised then the n_th host identity
is known. If all hosts are compromised, the originator is known.
Corollary 1:
Direct-transmission ARAs are vulnerable to an adversary that can
compromise any small subset of all hosts. This is done by sequentially
compromising the next host, and using that information to find
the identity of the next host after that. No amount of "random"
routing has any effect in this case, since the randomness is implemented
by each host, but each host is compromised before it makes the delivery.
Proposition 3:
Normal anonymous transmission (not ARA) is "unconditionally" secure
after *one* passage through an uncompromised host, assuming no
traffic analysis and no log files. (With log files, normal
transmission is as insecure as direct-transmission ARAs.)
Therefore, it seems like direct-transmission ARAs are much less
secure than normal anonymous transmission. For better security,
we must find some other ARA scheme.
A proposal: broadcast ARAs and Message Pools
- --------------------------------------------
All messages to a message pool are sent to all subscribers to the
pool. Messages to the pool are encrypted with the (pseudonymous)
public key of the recipient. The ARA can thus belong to any of
the subscribers to the pool. The connection between public keys
and subscribers is not maintained anywhere. The subscribers have
attempt decryption of messages marked with their pseudonyms.
Once the key of a subscriber is destroyed, it is not possible to
prove that any message was destined for that subscriber, affording
a last resort to a subscriber suspecting that an attack is in
progress.
Pools must have a large number of subscribers in case it is possible
to compromise the key of any particular subscriber. Pools can be
implemented as Usenet groups for a low-cost delivery medium. Each
pool should be geographically limited in order to further minimize
costs (the Distribution: header works well here). If costs are
minimized, the pools can be increased, affording better security.
For experimental, low-volume tests, mailing lists can be used.
- --
Miron Cuperman
participants (1)
-
miron＠extropia.wimsey.com