Re: Last Call: <draft-nottingham-safe-hint-05.txt> (The "safe" HTTP Preference) to Proposed Standard

[Mark Nottingham <[e-mail]>]

Hi Joe,

Thanks for the substantive comments. Responses (hopefully equal in substance) inline.

> On 14 Nov 2014, at 4:19 am, Joseph Lorenzo Hall <[e-mail]> wrote:
> 
> Signed PGP part
> 
> Hi, mnot has already heard the following concerns from us at CDT about
> this spec, but we want to make sure that these are part of the IETF
> last call comment record.
> 
> * The "Safe" preference is not only a preference but a signal.  It
>   signals user vulnerability; when activated, the header would signal
>   a user's potentially vulnerable status not only to site operators
>   who intend to reply in good faith, but to those that will operate in
>   bad faith and also to every intermediary on-path that could read the
>   preference request.
> 
>   Details about an Internet user's vulnerabilities should be treated
>   as sensitive information.  A broadcast signal that advertises a
>   user's content preferences or restrictions can signal her youth,
>   cognitive ability, relative media illiteracy, technological
>   inexperience, or another potential vulnerable status.  Because of
>   the risk that this information could be used to exploit immature or
>   inexperienced users, CDT generally cautions against widespread
>   identification of user vulnerability.  Obviously, sending such a
>   preference inside an encrypted connection removes concerns about
>   on-path observers, but not the more general concern with bad faith
>   endpoints or other embedded endpoints that may have other interests
>   (e.g., advertisers on a service may use this signal to profile
>   vulnerable populations).

This is definitely a concern, and one of the big reasons that it's restricted to one bit.

However, as pointed out, while it could indicate a vulnerable user, "safe" is also usable by others; anecdotally, many people turn "safe" mode on in search engines out of personal preference, not because they're children or otherwise vulnerable.

Additionally, "safe" is by no means a complete solution; it is not designed to be a filtering technology, nor to protect users from tracking / targeting; to attempt that it needs to be used in conjunction with other technologies (e.g., filters, ad blockers, tracker blockers, DNT -- although there is much debate about the effectiveness of these technologies as well).

The draft already mentions most of this, and I'm happy to strengthen that text if you think it will help.


> * Further, the ability for other intermediaries with access to the
>   request stream to insert the preference, potentially without notice
>   to the user, means that users may not even be aware that they are
>   broadcasting potentially sensitive information about themselves,
>   thus limiting their ability to take self-protective measures against
>   potential abuse.

Yes, this is possible.

The question, to me, is whether this potential abuse is outweighed by the potential benefit, and whether there are ways to mitigate the abuse.

Side story - When I was in Istanbul for IGF in September, I had a chat about "safe" with a well-respected researcher/advocate for online child safety. He was excited about it, because he had advocated something similar but much more bold to browsers in the past -- broadcasting the child's age bracket in requests.

Needless to say, that didn't happen, and I was surprised that someone with his knowledge and experience would consider putting such specific information in requests. I didn't ask at the time, but afterwards I surmise that he was implicitly relying on regulatory or legislative relief to assure that it wouldn't be abused.

"safe" is by nature much more conservative than this -- it only reveals one bit of information, and that bit is not specific to any one group of users. If someone is adding it to your requests (a situation that should becoming more uncommon, if numbers about HTTPS adoption are to be believed), that's happening within a legal context that can offer relief as appropriate.


> * As many of the comments in Last Call have identified, "Safe" content
>   in this specification is undefined. Because the proposal
>   (necessarily) lacks a definition of "safe", it is unlikely to be
>   useful to parents/guardians/users.  The lack of definition will
>   produce diverse and conflicting interpretations from content hosts
>   and providers, which can mislead users and their guardians, and may
>   invite abuse and confusion.
> 
>   The absence of guidance to websites wishing to participate in "safe"
>   content delivery will lead to varied and sometimes contradictory
>   results.  This could sow confusion and potential conflict among
>   participating platforms and website operators, and undermine the
>   utility of the specification.
> 
>   Moreover, users and their parents will have diverse expectations
>   about "safe" content.  These expectations will vary considerably
>   with users' age, as well as parent/guardians' cultural backgrounds.
>   Without a common understanding of what qualifies as "safe" content,
>   the expectations of users and their parent/guardians are likely to
>   be frustrated.  Of course, it is far outside the scope of a
>   technical specification to define a content-label like "safe".  But
>   because a standardized definition of "safe" content is unattainable,
>   the specification will have limited use as a tool for empowering
>   parents to regulate and guide their children's Internet use.

Again, "safe" is not a filtering technology; it is only pre-configuring the site's 'safe' mode if it has one, by the rules of that site. Whether or not that site's definition of 'safe' is appropriate to the user is a separate decision, one that might be enforced by a filter or some other mechanism.

In common deployments, I think it's best to view it as an extension of a Web filter *into* the site.

As above, I'm happy to expand / strengthen text here if you think it will help.

BTW, if the utility of the specification is indeed undermined as you point out, I think that's a largely self-correcting situation; lack of utility certainly hasn't stopped the IETF from making something a standard in the past.

Thanks again,

--
Mark Nottingham   https://www.mnot.net/