Re: [Bimi] Today's BoF

[Wei Chuang <[e-mail]>]

John,
Apologies for the delay as I was in transit.

*From: *John C Klensin <[e-mail]>
*Date: *Thu, Mar 28, 2019 at 5:01 PM
*To: * <[e-mail]>


>
> More into the IETF problem space, the presentation, as I
> understood it, seems to rely on technological ideas that have
> been tried and rather thoroughly demonstrated to not work or not
> be deployable in practice.


BIMI depends on DMARC/DKIM/SPF, and if the sender/receiver chooses to
validate the logo owner X.509.  Certainly the former has been deployed at
scale for email, and the latter web and other places at scale.


>   For starters, remember that we knew
> by the late 1980s what a technical solution that would have
> dealt with all of the key targets of the BIMI proposal and many
> others would look like: Sign message bodies, and optionally
> encrypt them, using a PKI that provided very strong assurance of
> the binding between the certificate, the authenticated identity
> of the signer, and with at least some connection to reputational
> information.  If that information did not check out on receipt,
> the message would either be discarded or would be delivered with
> very strong warnings. In particular, if brand identity
> information were sent with the message, the recipient could be
> assured of its authenticity by knowing where the message came
> from in much stronger ways than provided by SPF, DKIM, and/or
> DMARC.
>

On that I agree, that email sender based PKI aka S/MIME would provide much
strong i.e. per sender verifiable identiy.


>
> While the technology evolved somewhat, it never took off.   It
> would probably be safe to say that, with the exception of a
> rather few extra-paranoid or extra-privacy-sensitive individuals
> and some special applications, adoption has been about zero.
>

Agreed.


> There were several people at the BOF --in the room or remote--
> who could tell you at great length why that never happened
> (although we might have different theories about which elements
> were most important) but the bottom line is that, AFAICT, the
> model presented at today's BOF relies on either the the
> assumptions on that early idea


The BIMI proposal depends on different, domain based authentication
technologies that have been already deployed at scale.


> and would be likely to have
> deployment problems for similar reasons, or relies on variations
> on those assumptions that are so diluted as to make attacks by a
> determined party rather easy.
>

But agreed that using S/MIME would have been problematic, and that
constraint certainly guides the BIMI proposal.  My own informal polling was
the key management per sender is too difficult for most senders to deploy.
Even managing relatively few keys per domain as needed by DKIM and the
crypto involved is burdensome to many, hence the proposal still allows for
SPF.

My worry, as Dave Crocker's recent message on BIMI highlights, is that
these domain based authentication methods depend on the integrity of DNS,
and apparently there's now viable attacks on DNS (The Route 53/BGP hijack
is another).  Perhaps this proposal needs to take into account such DNS
attacks now rather than later.

-Wei


>     john
>
> --
> bimi mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/bimi
>