10 ms·

> Moving stuff around (from User-Agent to Sec-CH-UA-*) doesn't really solve much. That is, having to request this information before getting it doesn't help if

> Moving stuff around (from User-Agent to Sec-CH-UA-*) doesn't really solve much. That is, having to request this information before getting it doesn't help if sites routinely request all of it. I think this is sort of ignoring the whole point of the proposal. By making sites request this information rather than simply always sending it like the User-Agent header currently does, browsers gain the ability to deny excessively intrusive requests when they occur. That is to say, "sites routinely request all of it" is precisely the problem this proposal is intended to solve. There are some good points in this post about things which can be improved with specific Sec-CH-UA headers, but the overall position seems to be based on a failed understanding of the purpose of client hints.

marcosdumay 5y ago
Well, if the browsers can just deny those requests, then they can just drop the information entirely. (And they are dropping them from the UA.) From the two non-harmful pieces, one is of interest of all sites, and the other one has the implementation broken on Chrome, so sites will have to use an alternative mechanism anyway. If there's any value on the idea, Google can propose them with a set of information that brings value, instead of just fingerprinting people.
- Ajedi32 5y ago
  I think the idea is that there are some legitimate uses for UA information that they don't want to eliminate entirely, otherwise yeah they could just deprecate the User-Agent header and be done with it.
  - ocdtrekkie 5y ago
    I think most of the legitimate uses could be solved in a simple statement: Let users know whether the device is mobile or desktop, and then expect websites to send all of the logic to handle the rest client-side, so the server does not need to know. I'd love to see browser metrics being absolutely devastated as an analytic source: It just is used today as an excuse to only support Chrome.
    - blowski 5y ago
      Risk-based authentication can use a change in user agent as an increased risk factor.
      - Thiez 5y ago
        It could, but as someone who has spoofed user-agents in the past (primarily to get Chrome-only websites to cooperate) I would prefer if it wouldn't. If the baddies can snoop my https traffic or directly copy the auth cookies from my machine then also copying my user-agent isn't that big of a step for them. One might argue that detecting changes in user agents could be part of some kind of defense in depth strategy, but as a user I imagine I'm already so boned in that scenario that I doubt it would save me. So overall such a mechanism would bring me more inconvenience than security.
        blowski 5y ago
        That's the whole point of RBA, though. That two requests have the same user agent doesn't tell me much, but if you have two different user agents from two different IPs that may sound really risky (use case dependent, of course).
        ocdtrekkie 5y ago
        Unless someone is sitting at their desktop computer with their phone connection to 4G... Privacy initiatives will probably make some risk-based authentication tricks break, but they probably weren't robust methods anyways.
  - marcosdumay 5y ago
    Yes, I got that from your post. It's just that for Google, proposing it again with harmless content is very easy, but for anybody else to filter the bad content once the Google proposal gets accepted is almost impossible. (Although, if I was working on Firefox, I would just copy the most common data from Chrome, adjusting for those 2 fields that matter. That would create problems, but it's the less problematic choice.) So, no, it should be rejected. Entirely and severely. It doesn't mean that contextual headers are a bad practice, it's just that this one proposal is bad.
jsbdk 5y ago
>By making sites request this information rather than simply always sending it like the User-Agent header currently does, browsers gain the ability to deny excessively intrusive requests when they occur. Browsers can just not send a UA header
- tremon 5y ago
  I tried this. It breaks a surprisingly large number of sites (or perhaps not-so-surprisingly), and good luck trying to beat Google's captcha without a User-Agent header.
  - avian 5y ago
    Good luck trying to beat ReCaptcha if you're doing anything that puts you outside of the normal web browser behavior as imagined by Google's Algorithm. If User Agent Client Hints become the new normal, I'm sure anyone excessively denying requests will be flagged in the same way.
Svip 5y ago
> browsers gain the ability to deny excessively intrusive requests when they occur But Set-Cookie kind of proves what happen to that kind of feature. If at first sites gets used to be able to request it and get it, then the browsers that deny anything will simply be ignored. And then those browsers will start providing everything, because they don't want to be left out in the cold. That's what happened to User-Agent, that's what happened to Set-Cookie, and I can't see why it won't happen to Sec-CH-UA-*. Which the post hints at several times. Set-Cookie was supposed to have the browser ask the user to confirm whether they wanted to set a cookie. Not many clients doing that today. To be honest, I feel the proposal is a bit naïve if it thinks that websites and all browsers will suddenly be on their best behaviour.
- kijin 5y ago
  Yes, this looks like DNT all over again. Just another header that quickly becomes meaningless, wasting terabytes of bandwidth all over the world for no good reason.
  - wolverine876 5y ago
    DNT does nothing technically, but it has political power and that's where privacy happens to a great degree. When 70% of users say 'do not track me', it is hard to claim that they don't care about privacy.
    - m45t3r 5y ago
      Unless a big vendor (coff Microsoft coff) decides to enable it by default, them it becomes meaningless.
      - Analemma_ 5y ago
        It was meaningless from the beginning: DNT was always nothing but an Evil Bit. You’re getting mad at Microsoft for pointing out that the emperor had no clothes.
grishka 5y ago
Having to request it is a terrible idea to begin with. If I want to use different templates for mobile vs desktop, I need to know, on the backend, whether the device is a mobile device, and I need it on the very first request. Having to request these headers explicitly is an unnecessary complication that would slow down the first load. However it is nice that there's now a separate header that gives a yes or no answer on whether it's a mobile device.
- hypertele-Xii 5y ago
  Why would you need different templates for mobile/desktop? CSS is quite capable responding to any screen orientation.
  - jenscow 5y ago
    You're not wrong. However, there are times when CSS isn't enough. For example: - The Mobile vs Desktop design differences are too great. - The site was originally created without considering mobile, and retrofitting mobile support is unfeasible.
    - hypertele-Xii 5y ago
      Can you expand on the design differences?
  - grishka 5y ago
    Yes it is. Except you can't use the same markup for both because the input devices, and thus interaction paradigms, are so radically different. Mice are precise and capable of hovering over things, so it makes sense to pack everything densely and add various tooltips and popup menus. Touchscreens are imprecise and don't have anything resembling hovering, so UI elements must be large, with enough padding around them, and with menus appearing on click.
    - WorldMaker 5y ago
      Between CSS Flexbox and CSS Grid there shouldn't any reasons today that you can't handle 100% of those differences with the same markup and media stylesheets. (There's also obviously JS if you really must contort the HTML DOM to get what you want.)
jefftk 5y ago
Yes, I wish they would engage with how this fits into the rest of the Privacy Sandbox proposal (https://www.chromium.org/Home/chromium-privacy/privacy-sandbox https://www.chromium.org/Home/chromium-privacy/privacy-sandb...). My understanding is it's: 1. Move entropy from "you get it by default" to "you have to ask for it". 2. Add new APIs that allow you to do things that previously exposed a lot of entropy in a more private way. 3. Add a budget for the total amount of entropy a site is allowed to get for a user, preventing identifying users across sites through fingerprinting. Client hints are part of step #1. Not especially useful on its own, but when later combined with #3 sites now have a strong incentive to reduce what they ask for to just what they need. (Disclosure: I work on ads at Google, speaking only for myself)
- ocdtrekkie 5y ago
  I think pretty much all browsers and a lot of web platforms made it clear in their response to FLoC that everyone except Google (and Twitter, I guess?) considers Privacy Sandbox to be harmful as a whole.
  - jefftk 5y ago
    Objections to FLoC are basically about what should be included in #2. I don't understand why people would be opposed to #1 or #3 though?
    - ocdtrekkie 5y ago
      It's a fundamental disagreement on the very idea: Google's position is that it's okay for a website to know X amount of data about a user, you know, as long as it doesn't, in total, cross the creepy line. Everyone else's position is that if the data isn't required to operate, you don't need it. If we accept that the User Agent, as it is going to be frozen, is going to be served anyways to avoid breaking the legacy web, very little of this proposal adds value, and much of it adds harm. It isn't practical to move to not serving the User Agent, so any replacement for the data in it is pointless at it's very best. The frozen UA provides enough to determine if someone is mobile, the only real need for UA strings. And when most browsers are looking at reducing the tools for websites to fingerprint, Google is introducing new ones. So Firefox's position on Privacy Sandbox as a whole is pretty logical: If it's optional enough to be requested, why offer it at all? The entire premise of Privacy Sandbox is that it wants sites to have access to some amount of information about the user, and the position of every non-Google-browser is that they want to give sites as close to no data at all as possible. This is the core of the problem with a single company being legally permitted to operate a web browser and an ad company. Every single browser developer that doesn't own an Ads and Analytics suite is opposed to Privacy Sandbox.
1vuio0pswjnm7 5y ago
"By making sites request this information rather than simply sending it like the User-Agent header currently does..." This is also true with respect to SNI which leaks the domain name in clear text on the wire. The popular browsers send it even when it is not required. The forward proxy configuration I wrote distinguishes the sites (CDNs) that actually need SNI and the proxy only sends it when required. The majority of websites submitted to HN do not need it. I also require TLSv1.3 and strip out unecessary headers. It all works flawlessly with very few exceptions. We could argue that sending so much unecessary information as popular browsers do when technically it is not necessary for the user is user hostile. It is one-sided. "Tech" companies and others interested in online advertising have been using this data to their advantage for decades.
- billyhoffman 5y ago
  How would this work? SNI is sent by the client in the initial part of the TLS handshake. If you don't send it, the server sends the wrong/bad cert. The client could retry the handshake using SNI to get the correct cert but: - This adds an extra RTT, on the critical path of getting the base HTML, hurting performance. - A MITM could send back an invalid cert, causing the browser to retry with SNI, leaking it anyway (since we aren't talking about TLS 1.3 and an encrypted SNI). I suppose the client could maintain a list of sites that don't need SNI, like the HSTS preload list, but that seems like a ton of overhead to avoid sending unneeded SNI, especially when most DNS is unencrypted and would leak the hostname just like SNI anyways.
  - 1vuio0pswjnm7 5y ago
    "I suppose the client could maintain a list of sites that don't need SNI." That list would be much larger than the list of sites that do require SNI. Generally, I can determine whether SNI is required by IP address, i.e., whether it belongs to a CDN that requires SNI. Popular CDNs like AWS publish lists of their public IPs. I use TLSv1.3 plus ESNI with Cloudflare but they are currently the only CDN that supports it. Experimental but works great, IME. The proxy maintains the list not the browser. The proxy is designed for this and can easily hold lists of 10s of 1000s of domains in memory. That's more domains than I visit in one day, week, month or year. Is it not a question of whether this is possible. "How would this work". I have already implemented it. It works. It is not difficult to set up. Why this works for me and would unlikely work for others. I am not a heavy user of popular browsers, I "live on the command line". Installing a custom root certificate with appropriate SANs to suppress browser warnings is a nusiance that would likely dissuade others since they are heavy users of those programs. However I generally do not use those browsers to retrieve content from the web.

marcosdumay 5y ago

Well, if the browsers can just deny those requests, then they can just drop the information entirely. (And they are dropping them from the UA.) From the two non-harmful pieces, one is of interest of all sites, and the other one has the implementation broken on Chrome, so sites will have to use an alternative mechanism anyway. If there's any value on the idea, Google can propose them with a set of information that brings value, instead of just fingerprinting people.

Ajedi32 5y ago
I think the idea is that there are some legitimate uses for UA information that they don't want to eliminate entirely, otherwise yeah they could just deprecate the User-Agent header and be done with it.
- ocdtrekkie 5y ago
  I think most of the legitimate uses could be solved in a simple statement: Let users know whether the device is mobile or desktop, and then expect websites to send all of the logic to handle the rest client-side, so the server does not need to know. I'd love to see browser metrics being absolutely devastated as an analytic source: It just is used today as an excuse to only support Chrome.
  - blowski 5y ago
    Risk-based authentication can use a change in user agent as an increased risk factor.
    - Thiez 5y ago
      It could, but as someone who has spoofed user-agents in the past (primarily to get Chrome-only websites to cooperate) I would prefer if it wouldn't. If the baddies can snoop my https traffic or directly copy the auth cookies from my machine then also copying my user-agent isn't that big of a step for them. One might argue that detecting changes in user agents could be part of some kind of defense in depth strategy, but as a user I imagine I'm already so boned in that scenario that I doubt it would save me. So overall such a mechanism would bring me more inconvenience than security.
      - blowski 5y ago
        That's the whole point of RBA, though. That two requests have the same user agent doesn't tell me much, but if you have two different user agents from two different IPs that may sound really risky (use case dependent, of course).
        ocdtrekkie 5y ago
        Unless someone is sitting at their desktop computer with their phone connection to 4G... Privacy initiatives will probably make some risk-based authentication tricks break, but they probably weren't robust methods anyways.
- marcosdumay 5y ago
  Yes, I got that from your post. It's just that for Google, proposing it again with harmless content is very easy, but for anybody else to filter the bad content once the Google proposal gets accepted is almost impossible. (Although, if I was working on Firefox, I would just copy the most common data from Chrome, adjusting for those 2 fields that matter. That would create problems, but it's the less problematic choice.) So, no, it should be rejected. Entirely and severely. It doesn't mean that contextual headers are a bad practice, it's just that this one proposal is bad.

jsbdk 5y ago

>By making sites request this information rather than simply always sending it like the User-Agent header currently does, browsers gain the ability to deny excessively intrusive requests when they occur. Browsers can just not send a UA header

tremon 5y ago
I tried this. It breaks a surprisingly large number of sites (or perhaps not-so-surprisingly), and good luck trying to beat Google's captcha without a User-Agent header.
- avian 5y ago
  Good luck trying to beat ReCaptcha if you're doing anything that puts you outside of the normal web browser behavior as imagined by Google's Algorithm. If User Agent Client Hints become the new normal, I'm sure anyone excessively denying requests will be flagged in the same way.

Svip 5y ago

> browsers gain the ability to deny excessively intrusive requests when they occur But Set-Cookie kind of proves what happen to that kind of feature. If at first sites gets used to be able to request it and get it, then the browsers that deny anything will simply be ignored. And then those browsers will start providing everything, because they don't want to be left out in the cold. That's what happened to User-Agent, that's what happened to Set-Cookie, and I can't see why it won't happen to Sec-CH-UA-*. Which the post hints at several times. Set-Cookie was supposed to have the browser ask the user to confirm whether they wanted to set a cookie. Not many clients doing that today. To be honest, I feel the proposal is a bit naïve if it thinks that websites and all browsers will suddenly be on their best behaviour.

kijin 5y ago
Yes, this looks like DNT all over again. Just another header that quickly becomes meaningless, wasting terabytes of bandwidth all over the world for no good reason.
- wolverine876 5y ago
  DNT does nothing technically, but it has political power and that's where privacy happens to a great degree. When 70% of users say 'do not track me', it is hard to claim that they don't care about privacy.
  - m45t3r 5y ago
    Unless a big vendor (coff Microsoft coff) decides to enable it by default, them it becomes meaningless.
    - Analemma_ 5y ago
      It was meaningless from the beginning: DNT was always nothing but an Evil Bit. You’re getting mad at Microsoft for pointing out that the emperor had no clothes.

grishka 5y ago

Having to request it is a terrible idea to begin with. If I want to use different templates for mobile vs desktop, I need to know, on the backend, whether the device is a mobile device, and I need it on the very first request. Having to request these headers explicitly is an unnecessary complication that would slow down the first load. However it is nice that there's now a separate header that gives a yes or no answer on whether it's a mobile device.

hypertele-Xii 5y ago
Why would you need different templates for mobile/desktop? CSS is quite capable responding to any screen orientation.
- jenscow 5y ago
  You're not wrong. However, there are times when CSS isn't enough. For example: - The Mobile vs Desktop design differences are too great. - The site was originally created without considering mobile, and retrofitting mobile support is unfeasible.
  - hypertele-Xii 5y ago
    Can you expand on the design differences?
- grishka 5y ago
  Yes it is. Except you can't use the same markup for both because the input devices, and thus interaction paradigms, are so radically different. Mice are precise and capable of hovering over things, so it makes sense to pack everything densely and add various tooltips and popup menus. Touchscreens are imprecise and don't have anything resembling hovering, so UI elements must be large, with enough padding around them, and with menus appearing on click.
  - WorldMaker 5y ago
    Between CSS Flexbox and CSS Grid there shouldn't any reasons today that you can't handle 100% of those differences with the same markup and media stylesheets. (There's also obviously JS if you really must contort the HTML DOM to get what you want.)

jefftk 5y ago

Yes, I wish they would engage with how this fits into the rest of the Privacy Sandbox proposal (https://www.chromium.org/Home/chromium-privacy/privacy-sandbox https://www.chromium.org/Home/chromium-privacy/privacy-sandb...). My understanding is it's: 1. Move entropy from "you get it by default" to "you have to ask for it". 2. Add new APIs that allow you to do things that previously exposed a lot of entropy in a more private way. 3. Add a budget for the total amount of entropy a site is allowed to get for a user, preventing identifying users across sites through fingerprinting. Client hints are part of step #1. Not especially useful on its own, but when later combined with #3 sites now have a strong incentive to reduce what they ask for to just what they need. (Disclosure: I work on ads at Google, speaking only for myself)

ocdtrekkie 5y ago
I think pretty much all browsers and a lot of web platforms made it clear in their response to FLoC that everyone except Google (and Twitter, I guess?) considers Privacy Sandbox to be harmful as a whole.
- jefftk 5y ago
  Objections to FLoC are basically about what should be included in #2. I don't understand why people would be opposed to #1 or #3 though?
  - ocdtrekkie 5y ago
    It's a fundamental disagreement on the very idea: Google's position is that it's okay for a website to know X amount of data about a user, you know, as long as it doesn't, in total, cross the creepy line. Everyone else's position is that if the data isn't required to operate, you don't need it. If we accept that the User Agent, as it is going to be frozen, is going to be served anyways to avoid breaking the legacy web, very little of this proposal adds value, and much of it adds harm. It isn't practical to move to not serving the User Agent, so any replacement for the data in it is pointless at it's very best. The frozen UA provides enough to determine if someone is mobile, the only real need for UA strings. And when most browsers are looking at reducing the tools for websites to fingerprint, Google is introducing new ones. So Firefox's position on Privacy Sandbox as a whole is pretty logical: If it's optional enough to be requested, why offer it at all? The entire premise of Privacy Sandbox is that it wants sites to have access to some amount of information about the user, and the position of every non-Google-browser is that they want to give sites as close to no data at all as possible. This is the core of the problem with a single company being legally permitted to operate a web browser and an ad company. Every single browser developer that doesn't own an Ads and Analytics suite is opposed to Privacy Sandbox.

1vuio0pswjnm7 5y ago

"By making sites request this information rather than simply sending it like the User-Agent header currently does..." This is also true with respect to SNI which leaks the domain name in clear text on the wire. The popular browsers send it even when it is not required. The forward proxy configuration I wrote distinguishes the sites (CDNs) that actually need SNI and the proxy only sends it when required. The majority of websites submitted to HN do not need it. I also require TLSv1.3 and strip out unecessary headers. It all works flawlessly with very few exceptions. We could argue that sending so much unecessary information as popular browsers do when technically it is not necessary for the user is user hostile. It is one-sided. "Tech" companies and others interested in online advertising have been using this data to their advantage for decades.

billyhoffman 5y ago
How would this work? SNI is sent by the client in the initial part of the TLS handshake. If you don't send it, the server sends the wrong/bad cert. The client could retry the handshake using SNI to get the correct cert but: - This adds an extra RTT, on the critical path of getting the base HTML, hurting performance. - A MITM could send back an invalid cert, causing the browser to retry with SNI, leaking it anyway (since we aren't talking about TLS 1.3 and an encrypted SNI). I suppose the client could maintain a list of sites that don't need SNI, like the HSTS preload list, but that seems like a ton of overhead to avoid sending unneeded SNI, especially when most DNS is unencrypted and would leak the hostname just like SNI anyways.
- 1vuio0pswjnm7 5y ago
  "I suppose the client could maintain a list of sites that don't need SNI." That list would be much larger than the list of sites that do require SNI. Generally, I can determine whether SNI is required by IP address, i.e., whether it belongs to a CDN that requires SNI. Popular CDNs like AWS publish lists of their public IPs. I use TLSv1.3 plus ESNI with Cloudflare but they are currently the only CDN that supports it. Experimental but works great, IME. The proxy maintains the list not the browser. The proxy is designed for this and can easily hold lists of 10s of 1000s of domains in memory. That's more domains than I visit in one day, week, month or year. Is it not a question of whether this is possible. "How would this work". I have already implemented it. It works. It is not difficult to set up. Why this works for me and would unlikely work for others. I am not a heavy user of popular browsers, I "live on the command line". Installing a custom root certificate with appropriate SANs to suppress browser warnings is a nusiance that would likely dissuade others since they are heavy users of those programs. However I generally do not use those browsers to retrieve content from the web.