November 11, 2004

Opportunistic Cryptography is now Acceptable

Adam reports that Eric reports that the IETF has run its first meeting on opportunistic cryptography. Called "Better Than Nothing Security" the Internet protocol people are starting to get antsy about the number of attacks on unprotected sessions on the net.

And why aren't these sessions being protected by crypto? Because the original protocols called for a full-scale certificate authority signed key infrastructure. Nobody can be bothered with that, so they leave the links unprotected. Hence the notion that IPSec gives nothing because it isn't being utilised, and they may as well start accepting that crypto that is deployed is more useful than crypto that isn't deployed.

It is now some time past since we've had to argue the case. Crypto protocol people now seem to be comfortable that opportunistic cryptography - the glorified conceptual name for "wot SSH does" - is a valuable service.

There's still a little bit of loose naming to clear out. Adam points out that labelling opportunistic cryptography as "leaps of faith" is hardly helpful. Worse, it's just going to point more attention at the flaws in the certificate authority model.

For SSH and similar designs, we have one small step of faith at the beginning, In that if we don't check the certificate of the other node, *and* there is one of those vanishingly rare MITM attacks, then there exists a vulnerability. In contrast, we've pretty much established that certificate authority models call for two huge leaps of faith: firstly, that all CAs are the same, which isn't supported by any theory known to me, and secondly, that you can trust your CA because he is a trusted third party.

That last is now a highly questionable act of faith, in the light of Verisign's conflict of interest. Given that, the old label for self-signed certificates as "snake oil certs" is laughable, especially when phishing is considered. Self-signed certificates can make a huge difference to phishing, if the certificates were indexes into user-controlled information on the special sites. In contrast, certificate authority ones make no difference until we actually do something like adding branding to web browsers. Which then should we be calling snake oil?

Posted by iang at November 11, 2004 03:44 PM | TrackBack