And regarding the editorial comment of the submitted title:
- What it does is report only the most preferred language by default
- You can turn it off (i.e., just the language obfuscation) if you want to report all language preferences
I understand that it's easy to disable it, but it feels like a default that is strongly biased in favour of speakers of English and other widely spoken languages.
Maybe there should be a way to request all languages together and let the client pick whichever one it wants (I'm sure the text on a typical page when compressed, even times 50, is still negligible compared to the 50 MB of JavaScript frameworks it is probably pulling in) That would not sacrifice usability for privacy.
But here I see no added privacy in a normal setting if I do not use vpn, because 90% of IP addresses from Germany would report the same two languages. So only e.g. if I travel e.g. to Japan it makes me quite fingerprintable.
So I think the ideal would be if some entropy score could be displayed/predicted based on context (e.g. source or target address as above) and I could dynamically chose the trade-off between a bit of privacy and convenience.
The funny thing is that e.g. if you are in a country with a nonenglish speaking majority that has English as second language, just reporting either language assigns you to a smaller subgroup and makes no sense.
Something like
{ 0: de, 1: [en,fr], 2: [ru,es], 999: [zh,jp,...] }
where the keys are an arbitrary 'priority' score chosen by the server. 0 would be the original language, then in this example maybe 1 could be full human translation, 2 partial human translation, 999 machine translation.
The browser could keep its language preferences client-side and simply request their favourite language among the available ones.
We could replace 'breaks' with 'fixes', and we'd have the same kind of problem, but with the opposite bias: "Brave fixes language reporting in browser for more anonymity"