[cap-talk] SSL protection racket - Petname Tool

[Jed Donnelley]

On 2/26/2008 3:19 PM, Bill Frantz wrote:
> jed at nersc.gov (Jed Donnelley) on Tuesday, February 26, 2008 wrote:
> 
>>> Just create your own self-signed
>>> cert for the bank of your choice and away you go.
>> How does that get you into the middle?  All it does it to allow
>> you to set up another secure site - without paying any extortion
>> money to a protection racket.
> 
> You run a DNS poisoning attack, and get citi.com to come to you.
> You generate a self-signed cert for citi.com and use it on your
> server.

Such a self-signed cert would fail my Petname Tool trust
model (I assume?).  It would certainly be a different
public key than the one that I developed the trust in.

> You use the real citi.com server as a back end an you are
> now a man-in-the-middle.

My basic point is that I don't trust DNS and I don't trust
the network, but I do trust SSL to a certificate (private key)
that I've developed trust in through other means (introductions,
a past relationship).  It isn't the name in the certificate
that makes me trust it.  If that was my model all sorts of
confusion would be possible, naming ambiguities and conflicts.

My system (which, sadly, I must trust) picks the nonce and
does the handshake with the known trusted public key.  If
the handshake works then I trust the party at the other end
as much as my trust model has built up that trust (modulo
my trust in SSL).

Given that model, do you still believe that a self-signed
certificate or DNS failures or generally an untrustworthy
network (e.g. the Cyber Cafe example) results in my being
threatened by a man in the middle attack?

--Jed  http://www.webstart.com/jed/