SOME POSSIBLE ATTACKS ON RIPEM
------------------------------

This is a living list of potential weaknesses to keep your eyes open
for when using RIPEM for secure electronic mail.  It does not go into
great detail, and is almost certainly not exhaustive.  Obviously, many
of the weaknesses are weaknesses of cryptographically secured mail in
general, and will pertain to secure mail programs other than RIPEM.
It is maintained by Marc VanHeyningen <mvanheyn@cs.indiana.edu>.  It
is posted quartertly to a variety of news groups; followups pertaining
specifically to RIPEM should go to alt.security.ripem.

CRYPTANALYSIS ATTACKS
---------------------

- Breaking RSA would allow an attacker to find out your private key,
  in which case he could read any mail encrypted to you and sign
  messages with your private key.

  RSA is generally believed to be resistant to all standard
  cryptanalytic techniques.  Even a standard key (about 516 bits with
  RIPEM) is long enough to render this impractical, barring a
  huge investment in hardware or a breakthrough in factoring.

- Breaking DES would allow an attacker to read any given message,
  since the message itself is encrypted with DES.  It would not allow
  an attacker to claim to be you.

  DES has only 56 bits in its key, and thus could conceivably be
  compromised by brute force with sufficient hardware, but few agencies
  have such money to devote to simply read one message.  Since each
  message has a different DES key, the work for each message would
  remain high.  RIPEM 1.1 allows triple-DES to be used as an option;
  it is believed stronger than single-DES and should resist brute
  force attacks.

KEY MANAGEMENT ATTACKS
----------------------

- Stealing your private key would allow the same benefits as breaking
  RSA.  To safeguard it, it is encrypted with a DES key which is derived
  from a passphrase you type in.  However, if an attacker can get a copy
  of your private keyfile and your passphrase (by snooping network
  packets, tapping lines, or whatever) he could break the whole scheme.

  The main risk is that of transferring either the passphrase or the
  private key file across an untrusted link.  So don't do that.  Run 
  RIPEM on a trusted machine, preferably one sitting right in front of
  you.  Ideally, your own machine in your own home (or maybe office)
  which nobody else has physical access to.

- Fooling you into accepting a bogus public key for someone else could 
  allow an opponent to deceive you into sending secret messages to him
  rather than to the real recipient.  If the enemy can fool your
  intended recipient as well, he could re-encrypt the messages with
  the other bogus public key and pass them along.

  It is important to get the proper public keys of other people.
  The most common mechanism for this is finger; assuming the opponent
  has not compromised routers or daemons or such, finger can be 
  given a fair amount of trust.  The strongest method of key
  authentication is to exchange keys in person; however, this is
  not always practical.  Having other people "vouch for you" by
  signing a statement containing your key is possible, although 
  RIPEM doesn't have features for doing this as automatically as
  PGP.  RIPEM does generate and check MD5 fingerprints of public keys
  in the key files; they may be exchanged via a separate channel for
  authentication.

PLAYBACK ATTACKS
----------------

- Even if an opponent cannot break the cryptography, an opponent could
  still cause difficulties.  For example, suppose you send a message
  with MIC-ONLY (a PEM mode which does not provide disclosure protection)
  to Alice which says "OK, let's do that." Your opponent intercepts
  it, and now resends it to Bob, who now has a message which is
  authenticated as from you telling him to do that.  Of course, he may
  interpret it in an entirely different context.  Or your opponent
  could transmit the same message to the same recipient much later,
  figuring it would be seen differently at a later time.  Or the
  opponent could change the Originator-Name: to himself, register 
  your public key as his, and send a message hoping the recipient
  will send him return mail indicating (perhaps even quoting!) the
  unknown message.

  To defeat playback attacks, the plaintext of each message should 
  include some indication of the sender and recipient, and a unique
  identifier (typically the date).  A good front-end script for RIPEM
  should do this automatically (IMHO).  As a recipient, you should be
  sure that the Originator-Name: header and the sender indicated within
  the plaintext are the same, that you really are a recipient, and that
  the message is not an old one.  Some this also can and should be
  automated.  The author of this FAQ has made a modest attempt at
  automating the process of generating and checking encapsulated
  headers; the programs are included in the standard distribution in
  the utils directory.

LOCAL ATTACKS
-------------

- Clearly, the security of RIPEM cannot be greater than the security of
  the machine where the encryption is performed.  For example, under
  UNIX, a super-user could manage to get at your encrypted mail,
  although it would take some planning and effort to do something like
  replace the RIPEM executable with a Trojan horse or to get a copy of
  the plaintext, depending how it's stored.

  In addition, the link between you and the machine running RIPEM is
  an extension of that.  If you decrypt with RIPEM on a remote machine
  which you are connected to via network (or, worse yet, modem), an
  eavesdropper could see the plaintext (and probably also your
  passphrase.)

  RIPEM should only be executed on systems you trust, obviously.  In
  the extreme case, RIPEM should only be used on your own machine,
  which you have total control over and which nobody else has access
  to, which has only carefully examined software known to be free of
  viruses, and so on.  However, there's a very real trade-off between
  convenience and security here.

  A more moderately cautious user might use RIPEM on a UNIX workstation
  where other people have access (even root access), but increase
  security by keeping private keys and the (statically linked, of
  course) executable on a floppy disk.

  Some people will keep RIPEM on a multi-user system, but when dialing
  in over an insecure line, they will download the message to their
  own system and perform the RIPEM decryption there.  However, the
  security provided by such a mechanism is somewhat illusory; since
  you presumably type your cleartext password to log in, you've just
  given away the store, since the attacker can now log in as you and
  install traps in your account to steal your private key next time
  you use it from a less insecure line.  This will likely remain the
  situation as long as most systems use the rather quaint mechanism of
  cleartext password authentication.

  I find it nice to put a brief statement of how carefully I manage my
  security arrangement in my .plan next to my public key, so that
  potential correspondents can be aware what level of precautions are
  in place.  Some people use two keys, a short one which is not
  carefully managed for ordinary use and a longer one which is treated
  with greater care for critical correspondence.

UNTRUSTED PARTNER ATTACKS
-------------------------

- RIPEM's encryption will ensure that only a person with the private key
  corresponding to the public key used to encrypt the data may read the
  traffic.  However, once someone with that key gets the message, she
  may always make whatever kind of transformations she wishes.  There 
  exist no cryptographic barriers to a recipient, say, taking an
  ENCRYPTED message and converting it to a MIC-ONLY message, signed by
  you and readable by anyone, although RIPEM does not provide this
  functionality.  Indeed, the latest PEM draft I have seen specifically
  states that such transformations should be possible to allow
  forwarding functions to work.
 
  Including the recipients in the plaintext, as mentioned above, will
  make it possible for recipients of a redistributed message to be aware
  of its original nature.  Naturally, the security of the cryptography
  can never be greater than the security of the people using it.

TRAFFIC ANALYSIS ATTACKS
------------------------

- Some attacks are outside the scope of the PEM standard; traffic
  analysis is a prominent one of these.	 PEM does not prevent an enemy
  from potentially discovering who your traffic is being exchanged
  with and how often/lengthy these messages are.  This can be a
  problem for some people, though the potential for invasion of
  privacy may be more a collective than an individual one.  An
  interesting paper on a potential application of traffic analysis is
  mentioned below.

  The traditional way to prevent traffic analysis is to throw a lot of
  bogus traffic into the channel to obscure the real stuff; this could
  be done but would be rather detrimental to network load and bogus
  message recipients.  Trusted third-party re-mailers that handle
  aliases can help some, though aliases that are frequently used can
  still be analyzed (indeed, traffic analysis might determine which
  aliases go with which real people.)

  Interesting reference:
  Schwartz and Wood.  ``Discovering shared interest using graph
  analysis.''  To appear in CACM.
  
  Plain text version is in:
    ftp.cs.colorado.edu:/pub/cs/techreports/schwartz/ASCII/Email.Study.txt.Z
  Postscript version is in:
    ftp.cs.colorado.edu:/pub/cs/techreports/schwartz/PostScript/Email.Study