- From: Chaals McCathie Nevile <chaals@yandex.ru>
- Date: Wed, 07 Feb 2018 15:44:58 +0100
- To: public-privacy@w3.org, "Chris Lilley" <chris@w3.org>
First up, my apologies for not looking at this before. The functionality seems a really valuable enhancement to the Web, and I hope to find lots of cool new things making use of it. A couple of privacy concerns, and a couple of potential concerns led me to file two issues as described, and to ask some further questions below so I only make a fool of myself in one forum at a time... On Thu, 30 Nov 2017 17:39:57 +0100, Chris Lilley <chris@w3.org> wrote: > Audio WG requests privacy review of the Web Audio API > > * Specification URL: https://webaudio.github.io/web-audio-api/ > * GitHub repo: https://github.com/WebAudio/web-audio-api > > > There is a Security > and Privacy self- assessment appendix. > > > Ideally, we would like to move to CR by the end of 2017. Please > let us know if you would like more time to review. Review comments > should ideally be raised as issues on GitHub. I filed https://github.com/WebAudio/web-audio-api/issues/1487 which suggests that implementing this API in a voice-based system such as a screenreader or Voice Assistant increases the chance that a hostile site can emulate the native UI in some way. I note that this probably relies on either knowing enough about the user agent - including its particular sound characteristics - but that guessing at something common might be effective enough to justify the effort of a widespread attack. I also filed https://github.com/WebAudio/web-audio-api/issues/1486, noting that it seems the spec would allow "dolphin attacks" - using human-inaudible sounds to interact with sensor devices including voice assistants. (At the extreme end which I didn't consider is the alleged sonic attacks made on the US embassy in Cuba, apparently designed to cause physical harm. I do not know enough to check whether this is even potentially feasible with Web Audio). The same issue covers the case already identified for vibration, that an external sensor can detect the usage of the API potentially exposing physical location information. There is another issue I didn't file, because I am not sure about it. It seems plausible that it may be possible to use various filter effects to identify a user's hearing range, e.g. by overlaying different information at different frequencies. This information can be identifying both in the context of enabling further manipulation of a user cheers chaals -- Chaals is Charles McCathie Nevile find more at http://yandex.com
Received on Wednesday, 7 February 2018 14:46:36 UTC