Re: [MLS] Functional Definition of End-to-End Secure Messaging

[Dave Cridland <[e-mail]>]

Hi Alec,

On Thu, 6 May 2021 at 18:57, Alec Muffett <[e-mail]> wrote:

> What do you think, please?
>

As I said on Twitter, I'm hesitant to try to build a one-size-fits-all
definition of "secure". I think a much more useful first step (and one
where consensus is much easier to gain) would be to talk about different
threats and their potential mitigations and trade-offs, rather than
blanket statements that things are or are not secure in some absolute sense
which I'm unconvinced exists. "Secure" is not an absolute, and has to be
handled in context. MLS provides a (very) useful set of tools with which to
build various models of secure, including the consumer-grade personal
security you appear to be driving toward.

The primary objection I have here is that people will make the assumption
that any system that does not conform to your arbitrary definition of
"Secure" is, by inference, "Insecure".

As an example, consider §3.3. I currently write instant messaging
applications for the NHS (the UK's National Health Service, its
overwhelmingly dominant health provider) as well as others, for clinical
messaging. This obviously needs to be secure. Your stipulation is that
clinicians joining a particular group chat must not be able to see past
messages. This in turn means that clinicians must be ill-informed about the
patient's past, and therefore there is a heightened clinical risk to the
patient. I would argue that patient safety should be an outcome of an
applicable security stance. I think there is genuine risk involved in
providing people with a misleading single definition of "secure".

This is not to say that your definition is particularly "wrong"; it's not
capable of being wrong for the same reason that it's not capable of being
"right".

Dave.