3 ms·

I say it over and over. You can not completely anonymize data with any reliability. Please note the qualifier, many systems work for many vectors, but any suffi

I say it over and over. You can not completely anonymize data with any reliability. Please note the qualifier, many systems work for many vectors, but any sufficiently large dataset can be used to graph habits and correlate them. Maybe there is a safe way, but I put the onus of proving it on the person implementing it.

veeti 9y ago
RTFA: https://arxiv.org/pdf/1407.6981.pdf https://arxiv.org/pdf/1407.6981.pdf
digitalzombie 9y ago
> You can not completely anonymize data with any reliability. Well... there's actually a field for that. I forgot what they call that field because of how niche it is but my friend at google is doing just that. He said there are math theorem to prove that it's sufficiently anonymize. He gave an example of how Netflix competition with the data they gave researchers were able to deanonymize it. And his job was to prevent that at google. I can see why if you're trying to sell users data while maintaining privacy.
- kuschku 9y ago
  Mozilla currently uses Google Analytics for tracking, with "IP Anonymization" enabled. Which, according to Google’s FAQ, https://support.google.com/analytics/answer/2763052?hl=en https://support.google.com/analytics/answer/2763052?hl=en, just blanks out the last byte of the IP. Which is useless, because it still includes enough personalized data as to be completely and utterly reversible.
  - flyinghamster 9y ago
    Not to mention, if we're talking about IPv6 addresses, they often get allocated as /64 (or even /60 or /56) blocks. Such "anonymization" becomes useless at that point.
  - veeti 9y ago
    Google Analytics has nothing to do with this. As clearly linked in the mailing list, you can read the paper and source code for the client-side differential privacy tech used.
    - kuschku 9y ago
      Then check out https://github.com/mozilla/addons-frontend/issues/2785 https://github.com/mozilla/addons-frontend/issues/2785
- Vinnl 9y ago
  The linked Wikipedia page actually mentions that competition: https://en.wikipedia.org/wiki/Differential_privacy#Netflix_Prize https://en.wikipedia.org/wiki/Differential_privacy#Netflix_P...

veeti 9y ago

RTFA: https://arxiv.org/pdf/1407.6981.pdf https://arxiv.org/pdf/1407.6981.pdf

digitalzombie 9y ago

> You can not completely anonymize data with any reliability. Well... there's actually a field for that. I forgot what they call that field because of how niche it is but my friend at google is doing just that. He said there are math theorem to prove that it's sufficiently anonymize. He gave an example of how Netflix competition with the data they gave researchers were able to deanonymize it. And his job was to prevent that at google. I can see why if you're trying to sell users data while maintaining privacy.

kuschku 9y ago
Mozilla currently uses Google Analytics for tracking, with "IP Anonymization" enabled. Which, according to Google’s FAQ, https://support.google.com/analytics/answer/2763052?hl=en https://support.google.com/analytics/answer/2763052?hl=en, just blanks out the last byte of the IP. Which is useless, because it still includes enough personalized data as to be completely and utterly reversible.
- flyinghamster 9y ago
  Not to mention, if we're talking about IPv6 addresses, they often get allocated as /64 (or even /60 or /56) blocks. Such "anonymization" becomes useless at that point.
- veeti 9y ago
  Google Analytics has nothing to do with this. As clearly linked in the mailing list, you can read the paper and source code for the client-side differential privacy tech used.
  - kuschku 9y ago
    Then check out https://github.com/mozilla/addons-frontend/issues/2785 https://github.com/mozilla/addons-frontend/issues/2785
Vinnl 9y ago
The linked Wikipedia page actually mentions that competition: https://en.wikipedia.org/wiki/Differential_privacy#Netflix_Prize https://en.wikipedia.org/wiki/Differential_privacy#Netflix_P...