Comments inline.

- Phillip

> On Aug 21, 2025, at 9:39 PM, Murray S. Kucherawy <[email protected]> wrote:
> 
> On Wed, Aug 20, 2025 at 2:06 PM Phillip Tao <[email protected] 
> <mailto:[email protected]>> wrote:
>> > so if they're not going to sign it makes sense for them not to verify 
>> > either.
>> 
>> I don't think this necessarily follows from the preceding statement.
> 
> I think he meant: Client-to-client signing means the clients at each end need 
> to be able to do their bit, but if one end can't, then the other is moot.

I think "client-to-client" is meaningful in the context of confidentiality. 
When you're encrypting something, you're encrypting it _to_ someone, and 
expecting that only that specific party can decrypt it. Then, 
client-to-client/end-to-end has clearly different security characteristics than 
MTA-to-MTA.

However, in the context of signing, I don't see how "client-to-client" really 
means anything. When signing something, you're just establishing authenticity 
generally, and there's no way to specify (or limit) who is allowed to verify 
that authenticity.
> 
> Instead, signer-to-verifier operation where the signer is a domain MTA but 
> the verifier is an MUA is possible.  But you run into things like rotating 
> keys causing signatures to be invalid.  I think we talked about that 
> upthread. 

Yep, there are several problems. I think we should work on improving all of 
them.

> Also some operators use very short "x=" values as a way to mitigate replay 
> attacks (contrary to the standard), resulting in what you might think of as 
> false negatives. 

An explicit goal of DKIM2 is to prevent replay attacks, so I think this 
workaround should no longer be needed.

> There are probably other threats to this model I'm not thinking about without 
> more coffee.
> 
> Overall, I'd say client-side verifying might work for some or even most 
> cases, but I don't think I'd consider it to be very reliable.

It's decently (but not _very_) reliable today (with DKIM1), but I don't think 
we need to just take that as a given. There are specific reasons for that 
unreliability, and I think many of them can be solved. And I don't think we 
necessarily need to definitively solve _all_ of them, but we can solve some of 
them and take steps toward improving reliability for others.
>  
>> However, we've found that there are other policy decisions we want to apply 
>> to messages based on DKIM/DMARC status. For technical, security, and 
>> liability reasons, simply getting the DKIM/DMARC status from the 
>> Authentication-Results is not sufficient for this.
> 
> Can you explain why?  This is exactly what A-R was made to do.

The main issue with A-R results are that, for a MUA which is not limited to 
working with only a single mailbox provider controlled by the same 
organization, there's no (standardized/easy) way to establish trust that the 
A-R header was actually inserted by the MDA (i.e. section 7.1 of RFC 8601).
> 
> -MSK

_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to