Re: Formal objection to Encrypted Media Extensions progressing to Proposed Recommendation without greater user protection

On 08/02/2016 01:56 AM, Mark Watson wrote:
> Hi Harry,
>
> This may not be the right list for discussion of in-principle
> objections or the covenant proposal, but I wanted to quickly respond
> to a couple of mis-conceptions about our work below:
>
> On Mon, Aug 1, 2016 at 3:42 PM, Harry Halpin <hhalpin@w3.org
> <mailto:hhalpin@w3.org>> wrote:
>
>     [Note this is a formal objection as an individual in a private
>     capacity, not on behalf of my organization]
>
>     I'd like to fill a formal objection against Encrypted Media
>     Extensions progressing to Proposed Recommendation status without
>     adequate protection for users.
>
>     I believe that this work is so problematic given the well-known
>     and well-documented problems with DRM,  it should not happen as a
>     standard at all at W3C. That being said, for reasons which I do
>     not agree with and hope he reconsiders, the Director has approved
>     both the scope of the charter and the move of the Working Draft to
>     Candidate Recommendation.
>
>     If it does happen, then it seems given the well-documented
>     problems, some harm mitigation should be pursued.
>
>
> ​A large amount of effort has been devoted to this, yes.​
>  
>
>     The security research community has broad support for the EFF
>     covenant being a normative requirement [1]. However, it appears
>     consensus is not forthcoming on either EME itself or the EFF
>     covenant, with the Technology and Policy IG also failing to gain
>     any consensus for even further discussion (as it failed to even
>     get chartered).
>
>     To myself, the danger of EME is that over the last year suddenly
>     over millions of people had a content decryption module installed
>     without their explicit consent on their computer. For many users,
>     such as those of Firefox, the DCM was installed via a silent
>     update they had no control over. Such a content decryption module
>     can serve as both a technical security risk, as it puts a highly
>     privileged process in the user's computer
>
>
> ​I don't believe this is the case. The CDM - at least in Firefox's
> case - is highly sandboxed and the sandbox is open source code. So
> what the CDM can and cannot do on the user's computer is constrained
> and those constraints are open and verifiable.​ It certainly does not
> have access to the filesystem, for example and the handling of
> identifiers is very precisely constrained by the specification to
> avoid an increase in tracking surface.

I disagree. While of course sandboxing helps and should be encouraged,
research in the security community shows that browser sandboxing is
*never* perfect [1]. Thus, the security of the sandbox is *always*
relative to the code running in the sandbox. Also, I am not aware of any
actual formal verification or even thorough independent analyisis of
Firefox's sandbox regarding the CDM, or any other browsers. If you have
links, please share.


>
>  
>
>     outside their direct control by design, and as a legal risk
>     subjects any inspection of the DRM-enabled system on their
>     computer due to the anti-circumvention provision of the DMCA or
>     local equivalent. Thus, regardless of whether or not one agrees
>     with EME, it seems the *least* the W3C should do is warn the user
>     about the installation and activation of a CDM on their machine
>
>
> ​If such installation / activation exposed the users to any privacy or
> security risks, then yes, they should be asked to give consent.​
>
> To all of the points below, the User Agent is the user's agent and it
> is User Agent implementors who ultimately decide when consent /
> information is required to protect user security and privacy. If they
> choose not to do so, it is because they do not believe there are any
> additional security or privacy risks. We expect User Agent
> implementors to take this responsibility seriously and not to
> integrate with CDMs where they do not have sufficient information /
> insight to provide users with the guarantees they expect / deserve.
> The DMCA does not prevent CDM vendors sharing details of their product
> with their browser-implementing customers.

However, as *is already* the case with the Geolocation API, the spec can
quite easily specify that it's mandatory to ask users for consent before
usage of EME.  I think 'we expect user agent implementers to do the
right thing' is not protection enough for end-users.  The issue re the
DMCA is that it prevents sharing details of their CDM products to the
security research community and customers, who have shown again and
again to be able to discover flaws and security vulnerabilities that the
browser implementer community has missed [2].

>
> It is actually a *better* outcome if the CDM is so constrained that no
> consent is necessary (because there *are* no additional risks), than
> if there are additional risks and consent is therefore required.

As a CDM *always* creates additional risks (by design due to the
anti-circumvention clause of the DMCA), I believe that consent *must* be
necessary and so must disabling the CDM by default. As the arguments
offered here do not convince me and this change of the spec is clearly
within scope of the Working Group, I am not satisfied with the response
and my formal objected is still filed.

>
> ...Mark

cheers,
  
    harry

[1] http://arxiv.org/abs/1502.07373
[2] http://in.bgu.ac.il/en/Pages/news/CSRC-Chrome.aspx
>  
>
>     - and that the CDM should not be installed and EME should not be
>     activated without explicit user consent. Thus, in all
>     configurations, EME should be *de-activated by default.*
>
>     There is prior art in HTML for similarly powerful and
>     privacy-invasive features such as the Geolocation API [4]. Thus,
>     there should be no procedural problem in requiring user consent
>     and disabling EME by default in all browsers.  While it can be
>     argued many users will want to watch protected videos and will
>     turn them on, just as many users will want to use Google Maps with
>     Geolocation APIs, there does not seem to be a reasonable case for
>     having such
>     powerful and possibly dangerous features enabled by default.
>
>     Current language in the spec is so weak as it may not be enforced
>     and so EME does not have to be disabled by default: " User Agents
>     have some flexibility to determine whether consent is required for
>     a specific configuration and whether such consent may also apply
>     to other configurations. For example, consent to one configuration
>     may also imply consent for less powerful, more restricted
>     configurations. Equally, a denial of consent for one configuration
>     may imply denial of consent for more powerful, less restricted
>     configurations."
>
>     The only place where the spec [5] makes normative statements
>     around consent, but due to how DRM (i.e. Key Systems) are
>     designed, it is unclear how the user agent should interpret these
>     statements
>
>     "If a user agent chooses to support a Key System implementation
>     that cannot be sufficiently sandboxed or otherwise secured, the
>     user agent//should//ensure that users are fully informed and/or
>     give explicit consent before loading or invoking it."
>
>     "User Agents should ensure that users are fully informed and/or
>     give explicit consent before a Key System that presents security
>     concerns that are greater than other user agent features (e.g. DOM
>     content) may be accessed by an origin..."
>
>     " User agents must ensure that users are fully informed and/or
>     give explicit consent before Distinctive Identifier(s) are
>     exposed, such as in messages from the Key System implementation."
>
>     Although these statements show some progress towards trying to
>     mitigate the dangers of DRM, given that DRM systems work in
>     conjunction with anti-circumvention legislation that prevents a
>     developer from knowing whether or not they can sufficiently
>     sandboxed (or 'otherwise secured') and whether a Distinctive
>     Identifier is used by the DRM system, so the use of EME will
>     *always* present security concerns that are greater than the rest
>     of the user agent features involving the Open Web Platform.
>
>     Is the Working Group amendable to having EME and associated DRM
>     systems (i.e. "Key Systems") *normatively* disabled by default in
>     order to protect users?
>
>        cheers,
>            harry
>
>     [1]
>     https://www.eff.org/deeplinks/2016/03/security-researchers-tell-w3c-protect-researchers-who-investigate-browsers
>     [2]
>     https://en.wikipedia.org/wiki/Digital_rights_management#Opposition_to_DRM
>     [3]
>     https://en.wikipedia.org/wiki/Digital_rights_management#Shortcomings
>     [4]
>     https://www.w3.org/TR/geolocation-API/#implementation_considerations
>     [5] https://w3c.github.io/encrypted-media/
>
>