Re: [ietf-dkim] Not exactly not a threat analysis

Keith Moore wrote:

> > > Part of the idea that DKIM seems to propose is that more than one 
> > > party can potentially sign a message.  For instance, an author
> > > might sign a message, or a list might sign the same message. But
> > > different parties mean different things when they sign the message.
> > > If the author signs a message, it means "I wrote this".  If a list
> > > signs a message, it means "I sent this".
> >
> > But DKIM never gives an assertion of authorship (use PGP or S/MIME
> > for that).  Even if there is a valid signature that is associated
> > with the origination address, it means "the supposed author's domain
> > authorized this message".
>
> ...which is so vague as to be essentially useless.
>
> If the signature is not vouching for the authenticity of the content,
> and it's not vouching for transmission of the message to any particular
> recipient, what good is it?
Depends on what you mean by "authenticity".  Does the DKIM signature on 
this message mean that Jim Fenton wrote it?  No. Does it mean that the 
content is authentically from my domain?  Yes.
What good is it?  You could apply a whitelist, or eventually a 
reputation or accreditation service to tell you more about the domain 
that this message came from.
> What standard should a domain require before "authorizing" a message?
> That it's suitable for any recipient who might somehow find it in his
> mailbox?  (I doubt it)
That's up to the domain.  They might do sender authentication and just 
sign if they were assured that the sender is actually someone in their 
domain (and probably someone they could actually track down).  They 
might require that senders post a bond before they get there messages 
signed.  They might do an outgoing SpamAssassin check and only sign 
messages that look good.  There are lots of possibilities, and this is 
intentionally beyond the scope of the DKIM specifications.
> Should a domain penalize itself by refusing to "authorize" any message 
> that might, somehow, end up in the mailbox of someone  who didn't want 
> to receive it (and thus subject such messages to filtering even though 
> they were sent for a legitimate purpose)?  Or should a domain penalize 
> itself by "authorizing" all mail that it actually intended to send or 
> resend - even though that may harm its reputation?  (For an 
> organization that sends lots of mail, change "may" to "will".)
> Or should an enterprise have multiple domain names, and choose which 
> one to use as a signer based on the perceived risk to its reputation 
> associated with having it escape into the wild?
These are all possibilities.

> > This goes to what we have been very generically calling first-party
> > and third-party signatures.  The original submission of a message
> > would normally result in a first-party signature from the supposed
> > author's domain.  A mailing list would apply a third-party signature,
> > which can be distinguished by the fact that i= does not match the
> > originator's address.
>
> There are lots of reasons why i= might not match the originator's
> address, even if the originator actually authored the message.
Such as?  Clearly the signer can populate i= in any way it likes (and 
has a key that enables it to), but why wouldn't it sign the message on 
behalf of the OA at origination?
> > There are other circumstances where third-party signatures would be
> > applied as well, but I can't think of why it would be significant
> > whether the third-party signer is a mailing list, some other 
> > resender, or a greeting card or something.
>
> When evaluating a signature, I think the recipient generally cares 
> about one of two things: who wrote the message and who sent the 
> message to his mailbox.  The first tells him whether the message is 
> authentic; the second tells him who to blame for sending him a message 
> that he didn't want to receive.  In the second case, I don't think it 
> matters too much whether the signer is a mailing list or some other 
> kind of resender - except that in the case of a mailing list it might 
> not be feasible for the list to sign "I sent this message to X" for 
> each recipient X.
I think we have different views of what "authentic" means in this context.

-Jim
_______________________________________________
ietf-dkim mailing list
http://dkim.org