Re: [ietf-privacy] Is there an official working definition for Privacy Online?

[Nick Doty <[e-mail]>]

On May 5, 2016, at 7:53 AM, Alissa Cooper <[e-mail]> wrote:
>> On May 2, 2016, at 2:29 PM, Christian Huitema <[e-mail]> wrote:
>> 
>> On Sunday, May 1, 2016 4:12 PM, Dave Crocker wrote:
>>> 
>>> If the term is to be a non-technical and vague reference, then let's stop
>> using it
>>> as if it were a technical term.  Philosophical, academic and social terms
>> are
>>> fine; the problem is when we use them as if they pertained to technical
>>> specifics.
>> 
>> Well, we do use the term "security" liberally, don't we? It is certainly
>> just as vague, but it is useful as a section header. It encourages protocol
>> designers to be concerned with the broad issue of security attacks. I think
>> that we have consensus that protocol designers should also be concerned with
>> the broad issue of privacy attacks.
> 
> +1. If people want to consider privacy as a heading under which we group a bunch of different kinds of attacks, that works perfectly well I think.
> 
> Rather than spending a lot of time to try to find a magical two-sentence definition that everyone can agree on (which I doubt is feasible), I think the time would be better spent on refining how we define the set of attacks and mitigations against them, building on or fixing what’s in RFC 6973, possibly turning bits of that into a BCP, etc. The two sentences will not be directly actionable no matter what they say, whereas a comprehensive threat model and mitigations suite could be.

+1. RFC 3552 doesn't provide a two-sentence definition of "security", but points out that the definition is not simple, but instead a common grouping of properties and threats, which get more detailed definitions in subsequent sections. Similarly, RFC 6973's terminology section does provide technical definitions of terms like unlinkability, fingerprinting, etc.

I do tend to agree with the conclusion that as a result we shouldn't be using "privacy" as a technical term on its own. Sentences in specs of the form "Feature X undermines privacy" or "Feature Y provides privacy to the end user" are either inappropriate, or more likely, incomplete. Instead: "Feature Y supports privacy by providing unlinkability of traffic between requests".

It might still be useful to refine a short definition that can be cited to speed up conversations on privacy at IETF or help in scoping work; NIST, for example, has been trying to come up with privacy engineering objectives analogous to the C-I-A triad for security.

Cheers,
Nick