Browser Fingerprinting | Hacker News Reader

Browser Fingerprinting

sites.google.com

146 pointsdedalus43 comments

Mozilla and Tor devs are currently working on upstreaming some Tor Browser patches[0] that reduce fingerprinting surface. E.g. cookie isolation based the domain shown in the URL bar[1] and making canvas readback opt-in[2]

[0] https://wiki.mozilla.org/Security/Tor_Uplift/Tracking [1] https://bugzilla.mozilla.org/show_bug.cgi?id=1299996 [2] https://bugzilla.mozilla.org/show_bug.cgi?id=967895

What about these? https://www.browserleaks.com/firefox https://bugzilla.mozilla.org/show_bug.cgi?id=863246 https://bugzilla.mozilla.org/show_bug.cgi?id=903959

Hence the "E.g."

https://panopticlick.eff.org/

My font set was completely unique, which I partly expected. However, my canvas fingerprint was also completely unique, which did surprise me.

Any idea where we can find the most common subset of system fonts?

Isn't canvas fingerprint linked to your window size, if you maximize your browser, does that help?

It's quite surprising that my disposable virtual machine's firefox in Qubes OS is unique among all tested so far...

Why is that surprising? That's going to be a unique fingerprint in most datasets. The average user is not using Qubes, Firefox or Linux.

Qubes isn't designed to blend into all other users; it's designed to make you unlinkable to your other Qubes instances.

A better experiment would be to run the test in two of your Qubes instances. Hopefully they will both be unique.

Isn't Qubes OS just a hypervisor? The hosted OSes shouldn't leak that they are running under Qubes.

Yeah, but the point still stands most people are not running all the unique programs within the guest OSes (Linux, Firefox, Tor, ...).

But granted it may be surprising there was no collision with another HN user.

Note that this is very outdated. Modern fingerprinting techniques can identify you much more accurately than panopticlick can.

In Firefox, you can control a lot via the `about:config` directives. You can put all that into a `user.js` for portability and reference, like this:

https://github.com/delight-im/Secure-Firefox

But unfortunately, there are so many things that can contribute to identification of your browser, and many of them cannot reasonably be blocked. The important section of the Chromium list is "(2) Machine-specific characteristics". You need most of that, so you can't block or hide it.

Privacy is not easy. Not to mention the situation for non-technical users.

Hidden prefs are hidden for a reason and can be a foot-gun. The warning when opening about:config is no exaggeration, and copying-and-pasting a bunch of hidden prefs from an untrusted source is not a good idea.

For instance:

> user_pref("network.cookie.cookieBehavior", 1);

If I remember correctly, this may break some sites.

> user_pref("dom.battery.enabled", false);

Questionable value for privacy / fingerprinting. If you plug-in your computer or pull the plug (a pretty common scenario), the site trying to identify you will suddenly think you're a different users.

> user_pref("general.buildID.override", "20100101");

Pointless since 20100101 is already the default used in the user agent string. Makes me think this list is severely outdated or the author doesn't know what he's doing.

> user_pref("browser.newtabpage.enabled", false);

Little impact on privacy (is this about the "somebody opens a new tab in my profile and sees what sites I frequently visit" scenario?), no impact on fingerprinting. The readme says "Impacts on user experience should be minimized, erring on the side of convenience." How does disabling the new tab page satisfy that?

> The warning when opening about:config is no exaggeration

If you feel uncomfortable in that section, feel free to avoid touching it. But if you know what you do, there are zero problems. Want to change your user agent? Do it there. No problems. The warning is for your average non-technical user, justifiably.

Well, perhaps you need to read up on those a little bit again.

> If I remember correctly, this may break some sites.

Sure, everything may break some sites. But this is about blocking third party cookies, which blocks not a single site I know (Do you know a popular one?) while being one of the most effective measures against tracking (which uses fingerprinting) out there.

> Questionable value for privacy / fingerprinting. If you plug-in your computer ...

Sure, agreed for privacy.

> Pointless since 20100101 is already the default used in the user agent string

Yes, since late 2015 only, and keeping a default hardly does any harm. By the way, try outputting `navigator.buildID` in your JavaScript console -- oops, not even this setting helped. The real ID is right there. So it's not as easy as you might think.

> is this about the "stranger opens a new tab in my profile and sees what sites I frequently visit" scenario?

No. If I remember correctly, the "new tab" page is about the ads.

> If you feel uncomfortable in that section, feel free to avoid touching it.

I feel very comfortable touching hidden prefs in Firefox. That's not my point. My point is that random users stumbling upon your github page may not know what they're doing.

> Sure, everything may break some sites. But this is about blocking third party cookies, which blocks not a single site I know (Do you know a popular one?)

I quickly googled this for you, found this:

https://bugzilla.mozilla.org/show_bug.cgi?id=849948

https://news.ycombinator.com/item?id=5272069

https://getsatisfaction.com/getsatisfaction/topics/bad_user_...

https://planapple.uservoice.com/knowledgebase/articles/50965...

> No. If I remember correctly, the "new tab" page is about the ads.

The new tab page displays frequently visited sites. In new profiles without a browsing history, it would also suggest sites, some of which would be "sponsored" (i.e. ads). If I remember correctly, Firefox would pick the ads locally without leaking information about you. So what's the privacy impact? Also, it doesn't even do that anymore: http://arstechnica.com/information-technology/2015/12/firefo...

> I quickly googled this for you, found this

Right, "1292 days ago". Now that we've just talked about information that is out-of-date.

> Blocking 3rd party cookies breaks surprisingly few things

Agreed. A comment from the post that you linked to.

Disqus being one of the few exceptions that don't work (because it's embedded as a third-party on first-party sites), personally, I can live with that. Everybody has to decide for themselves. If some setting doesn't work, it can simply be undone.

> Disqus being one of the few exceptions that don't work (because it's embedded as a third-party on first-party sites),

Just like e.g. Facebook Login and github accounts, right?

> personally, I can live with that. Everybody has to decide for themselves. If some setting doesn't work, it can simply be undone.

Except that the list that you promoted here has no detailed documentation and doesn't put users in the position to make an informed decision.

Facebook login, Google login, etc. still work (at least with most implementations), since they redirect to the SSO site, which is what every reasonable site should do, instead of using an iframe/embed.

Good luck finding something that considerably improves your privacy while breaking 0.000% of modern sites.

I'm not blaming you for the fact that disabling third-party cookies may break some sites. The problem is that you provided a long list of preferences privacy-sensitive users allegedly should set, without explaining what these prefs do let alone warning about the possible negative impact.

  > Pointless since 20100101 is already the default used in the user agent string
  
  Yes, since late 2015 only

Actually it was 2010 (hence 20100101) for Firefox branded builds:

https://bugzilla.mozilla.org/show_bug.cgi?id=591537

And 2013 for all builds (I happen to have fixed this bug):

https://bugzilla.mozilla.org/show_bug.cgi?id=728773

>> user_pref("dom.battery.enabled", false);

In current Firefox this is AFAIK whitened exactly to prevent fingerprinting. Disabling it will increase your footprint because it's not the default.

There is one more called Firefox Debloat: https://github.com/amq/firefox-debloat/

... though it hasn't been recently updated.

We are developing `gngr` in the belief that privacy should be engineered into the browser, not worked around:

https://github.com/UprootLabs/gngr

A new browser, neat!

Offtopic: I am looking for a cross-platform browser written in Go(lang) even though the GUI stuff is just getting started there.

  Privacy is not easy.

Please don't adopt this tagline. This quick catchphrase is the shortest path to lies.

This kind of thinking basically helps powerful people erode the idea that they are not entitled to anybody's personal details.

Privacy is as easy as not shooting yourself in the foot.

You just have to not point the gun at your foot.

It is easy not to shoot yourself in the foot if you do not have a gun.

Privacy and convenience are at two ends of a spectrum. Commerce likes to exploit this by making things more "convenient", generally at the cost of privacy. Sadly, the majority of consumers care far more about convenience than the often hidden costs of lack of privacy.

That's great.

It's still a bullshit catchphrase, empty as any buzzword.

Once talked with over a beer with a web performance head of company X, he said that you can use the web performance profile - without looking at any other browser identifiers - to pretty accurately do browser fingerprinting. When you then look additionally at the interaction speed within an webapp you can even profile different users.

http://valve.github.io/fingerprintjs2/

In one of the sections:

> for example, TCP/IP traffic will often reveal high-resolution system uptime data through TCP timestamps.

Is this something that's on by default? Dependent on OS? Not deeply familiar with the details of the TCP stack, but have never hard of something like this being exposed via TCP requests!

Did some research - it's off by default in recent macOS versions. Configurable via the "sysctl net.inet.icmp.timestamp" flag.

http://superuser.com/questions/680200/os-x-how-to-make-it-re...

I just tried this out, and it seems that "system fonts" are the biggest identifier of mine.

Any idea where I can find the most common subset of system fonts?

I'd love a settings checkbox saying something like "Disable user-installed fonts".

Is there a way to easily change (some) browser plugins every five minutes, so fingerprinting becomes useless?

You could load your browser through Selenium and explicitly choose which plugins to load but I wouldn't call it easy. Definitely not accessible to the average user.

Perhaps somebody could write a plugin which changes identity (as seen by the fingerprinter) every so often.

On Chrome you can use this[0] user-agent spoofer(one example) It will randomize your user agent for each session, thus changing your fingerprint.

I'm sure there are various other extensions that do similar things that randomize the parameters of whatever is used to fingerprint you.

[0] https://chrome.google.com/webstore/detail/random-user-agent/...

That will set off some fraud detection on sites as there will be mismatch between user agent and the browser being used. You can easily profile the DOM and Javascript features to figure out what browser a person is using.

On Firefox there is: https://addons.mozilla.org/firefox/addon/random-agent-spoofe...

Plugins are a very small aspect of fingerprinting and there's really nothing you can do to stop it other than build a browser specifically designed to stop it.

Using multiple VMs with diverse connectivity, one can sidestep all "Explicitly assigned client-side identifiers" and "Machine-specific characteristics". For evading WebGL, VM OS diversity is essential. It's harder to obscure "User-dependent behaviors and preferences", especially mouse and keyboard stuff. But it helps to switch keyboards, and between using mouse and touchpad.

"users who extensively tweak their settings from the defaults may be actually making their browsers considerably easier to uniquely fingerprint" ...interesting comment in section, "User-dependent behaviors and preferences"

I don't see evercookie mentioned: http://samy.pl/evercookie/ which implements something similar

Internet Tracking Has Moved Beyond Cookies

https://fivethirtyeight.com/features/internet-tracking-has-m...

Reply on news.ycombinator.com