sign and/or encrypt SAML assetions, hack MITM

[Cantor, Scott]

> Testing for this on a larger scale isn't trivial (and might include
> legal aspects) but an activity within GÉANT has recently started to
> look into this:

My only comment was going to be that, yeah, point of fact this is very hard to test other than to...actually test it. People just assume their vendors are doing the right stuff....I'm here to tell you they ain't.

You have multiple, likely critical, apps right now that are on fire and you simply don't know it.

Encryption's main value is not the confidentality, it's the fact that encrypting makes it much harder for those apps to get this wrong, and for stolen messages to be played back to other broken sites.

I'm moderately surprised that OIDC didn't simply take the path SAML Artifacts did...the whole point of them was to allow people to not do signature checks. That was the win, in exchange for adding the back channel.

I guarantee that a lot of OIDC clients aren't doing those checks, but they have the additional constraints when the implicit flow isn't used, and whatever their public reasoning was, the real reason for denigrating the implicit flow was they damn well know it.

-- Scott