Google AMP doesn't show Google on the page. Google is pushing for the URL to show the origin site's URL instead of Google[2].
If an attacker poisons a nytimes.com article served by Google AMP, how does a browser's domain blacklisting help? Block google? Block nytimes.com? Neither makes sense.
1. https://web.archive.org/web/20050401090916/http://www.google...
2. https://9to5google.com/2019/04/18/apple-mozilla-google-amp-s...
example.com generates a content bundle and signs it. Google.com downloads the bundle and decides to mirror it from their domain. Your browser downloads the bundle from google.com, and verifies that the signature comes from example.com. Your browser is now confident that the content did originate from example.com, and so can freely say that the "canonical URL" for the content is example.com.
Malicious.org does the same thing, and the browser spots that malicious.org is blocked. At this point it doesn't matter if the content came from google.com, because the browser knows that the content is signed by malicious.org and so it originated from there.
Hope this helps clarify. Obviously blacklisting isn't a great security mechanism; my point is just that signed exchanges don't really open any NEW vectors for attack.
Imagine that example.com builds the bundle by pulling data from a database. If an attacker can find a way to store malicious content in that database (stored XSS) and that content ends up in a signed bundle that Google AMP serves (similar to cache poisoning) then users will see malicious content. When the stored XSS is removed from the database, Google AMP may continue to serve the malicous signed bundle. So an extra step may be needed to clear the malicious content from Google AMP.
How exactly the attacker influences the bundle is going to be implementation dependent, so some sites may be safe while others are exploitable.
If, purely as a hypothetical, Russian operatives got a credible propaganda story posted on the NYT website 24 hours before the November elections, and an AMP-hosted version of it stayed live long after the actual post got removed from nyt.com, I'd certainly call that "malicious". Of course, just like archive.org, I suspect that in a case as high-profile as that, you'd see a human from the NYT on the phone with a human at Google to get the cached copy yanked ASAP, but maybe on a slightly smaller scale the delay could be hours-to-days, which is bad enough.
This expiration can also never be set more than 7 days in the future.
"Signing a bad response can affect more users than simply serving a bad response, since a served response will only affect users who make a request while the bad version is live, while an attacker can forward a signed response until its signature expires. Publishers should consider shorter signature expiration times than they use for cache expiration times."
The serving nodes are not necessarily under control of a well intended party that complies with upgrade requests.
Sure, you could move it somewhere else and have it show up in the address bar the same, but the actual URL has changed and you need to somehow get the new URL into people's hands. And ultimately you've centralized a lot of websites under a smaller number of service providers which, before, would have been on their own domains.
Isn't signed exchanges basically CDN's without having to setup DNS? It's in theory no different than using CloudFlare to serve your content, except any CDN can just serve it without giving them access to your domain.
How so?
AMP is a scourge. It's a bad idea being pushed by bad actors.
I think what Spivak is saying though is right. If we could move from location addressing (dns+ip) to content-addressing , but not via the AMP cache servers, in general, anyone could serve any content on the web. Add in signing of the content addressing, and now you can also verify that content is coming from NYTimes for example.
Also, I'd say that the internet (transports, piping, glue) is decentralized. The web is not. Nothing seems to work with each other and most web properties are fighting against each other, not together. Not at all like the internet is built. The web is basically ~10 big silos right now, that would probably kill their API endpoints if they could.
I don't think this should be shoehorned into the URL bar or into some meta info that no one ever reads hidden behind some obscure icon.
It actually makes perfect sense in Doublespeak. /s
But AMP is a much narrower technology, I’d imagine only Google would be able to impersonate other websites, essentially centralised as you say. The generic idea would just be a distraction to push AMP.
Everything would be so much better if the original websites were not so overloaded with trackers, ads and banners, then there would be no need for these “accelerated” versions.
Could there be net-neutrality-like questions in all this as well?
Create a new “original URL” field or something.
What you're saying would be described as distributed... Not decentralized.
Stated another way, with a typical CDN setup the user has to trust their browser, the CDN, and the source. With signed exchanges we're back to the minimal requirement of trusting the browser and the source; the distributor isn't able to make modifications.
What a bummer.
(Disclosure: I work for Google, speaking only for myself)
And by hovering, or one-clicking, a popup could show both the distributor's address (say, CloudFlare), and the content's/publisher's address (say, NyT)?
https://blog.cloudflare.com/keyless-ssl-the-nitty-gritty-tec... is a thing now.
To put it simply, Cloudflare still controls the content. The proposal here would avoid that, by allowing Cloudflare to transmit only pre-signed content.
Google controls the AMP project and the AMP library. They can start rewriting all links in AMP containers to Google’s AMP cache and track you across the entire internet, even when you are 50 clicks away from google.com.
Really? Could you publish how you are inspecting an unknown program to determine if it exhibits a specific behavior? There are a lot of computer scientists interested in your solution to the halting problem.
Joking aside, we already know from the halting problem[1] that it you cannot determine if a program will execute the simplest behavior: halting. Inspecting a program for more complex behaviors is almost always undecidable[2].
In this particular situation where Google is serving an unknown Javascript program, a look at the company's history and business model suggests that the probability they are using that Javascript to track use behavior is very high.
def divisors(n):
for d in range(1, n):
if n % d == 0:
yield d
n = 1
while True:
if n == sum(divisors(n)):
break
n += 2
print(n)
Also, if I execute a program and it does exhibit that behaviour, that’s a proof right there.
The same kind of analysis can be applied to Google’s scripts: look what data it collects and where it pushes data to the outside world. If there are any undecidable problems along the way, then Google has no plausible deniability that some nefarious behaviour is possible. Now, whether that is a practical thing to do is another matter; but the halting problem is just a distraction.
Tracking doesn't require reading any of your data. All that is necessary is to trigger some kind of signal back to Google's servers on whatever user behavior they are interested in tracking.
> or communicate with third parties
Third parties like Google? Which is kind of the point?
> [example source code]
Of course you can generate examples that are trivial to inspect. Real world problems are far harder to understand. Source is minified/uglified/obfuscated, and "bad" behaviors might intermingle with legitimate actions.
Instead of speculating, here is Google's JS for AMP pages:
https://cdn.ampproject.org/v0.js
How much tracking does that library implement? What data does it exfiltrate from the user's browser back to Google? It obviously communicates with Google's servers; can you characterize if these communications are "good" or "bad"?
Even if you spent the time and effort to manually answer these questions, the javascript might change at any time. Unless you're willing to stop using all AMP pages every time Google changes their JS and you perform another manual inspection, you are going to need some sort of automated process that can inspect and characterize unknown programs. Which is where you will run into the halting problem.
Be cool if you did ;)
This has nothing to do with the halting problem because that is concerned about for all possible programs not some programs.
We obviously know if some programs halt.
while true: nop
X = 1
Y = X + 2
More complex behaviours can be easier. Neither of my programs there make network calls.
Likewise, AMP pages are mostly accessed from Google search that's already tracked.
Publisher hosted copies are in the pipeline, as I referenced in the parent comment. My choice of verbiage was a bit confusing it appears.
AMP's documentation seems to indicate that the LTS is stable only for one month (new features released via the same URL each month), and so is not compatible with SRI (see https://github.com/ampproject/amphtml/blob/master/contributi...)
You can specify a version (ie, https://cdn.ampproject.org/rtv/somenum/v0.js), but the AMP validator complains about that.
Yes, and? What’s your point? It’s actually a security weakness to include third party JS. The whole thing runs on trust.
"AMP HTML documents MUST..."
"The AMP runtime is loaded via the mandatory <script src="https://cdn.ampproject.org/v0.js"></script> tag in the AMP document <head>."
Do a whois on ampproject.org:
"Registrant Organization: Google LLC Registrant State/Province: CA Registrant Country: US Admin Organization: Google LLC"
Note that jQuery, as mentioned in some GP comment has no such requirement. Google AMP is quite unique in this regard. This is NOT some general CDN type issue. Also...agreed, WTF is "open intent"?
They cannot be allowed to become the gatekeeper for the web.
It seems to me, a lot of the security concerns come from the requirements to make pages served live and pages served from bundles indistinguishable to a user - a requirement that really only makes sense if you're Google and want to make people trust your AMP cache more.
I'd be excited about an alternative proposal for bundles that explicitly distinguishes bundle use in the URL (and also uses a unique origin for all files of the bundle).
https://m.youtube.com/watch?v=gqGEMQveoqg
(Google Tech Talk from Van Jacobsen on CCN many years ago)
I don’t support ads but I also don’t support Google serving a version of the page that steals money from content creators. So, therein lies the problem: choice.
I can imagine a future where amp is ubiquitous and Google begins serving ads on amp content. Luckily, companies have to make money and amp is not in most people’s or company’s best interests.
If amp was opt-in only, this would be much more ethically sound.
Google has never injected ads into any cache served AMP document (technically if the publisher uses AdSense, this is false, but that's not the point you are making).
It's difficult to follow what definition of theft is being suggested. The cache does not modify the document rendering, it's essentially a proxy. In a semantic sense, this is no different than your ISP delivering the page or your WiFi router.
Perhaps that's a great thing to do, but it's not something to do quietly.
https://www.techradar.com/uk/news/google-is-phasing-out-thir...
https://blog.chromium.org/2020/01/building-more-private-web-...
I'm going to copy paste my older comment on this:
I find their "removing 3rd party cookies will incentivise businesses to rely on fingerprinting" discourse dangerous.
It implies that other browser vendors (Mozilla, Safari/WebKit, new Edge) are in fact making the Web a more dangerous place.
I believe it's dangerous because it creates a harmful, unproductive PR narrative—people might just assume this is a true statement, without learning about both sides of the problem. I'm not trying to strip anyone of agency, I just don't think most of my friends would have time to research this topic and might decide to follow the main opinion instead.
The answer I'd like to hear: Yes, it does push some actors towards fingerprinting, but preventing fingerprinting should be dealt with regardless. Changes should happen both on legislative and browser-vendor level.
Cloudflare allow using of same domain to use AMP. In this case, content is served from Cloudflare CDN.
As I didn't know about this addon, thanks for sharing it.
[1] https://addons.mozilla.org/en-US/android/addon/amp2html/
One should never forget that at a certain point, Google will likely invoke the looser's argument ("protect you from terrorists and pedophiles") to require proof of identity prior to granting access to any resource or service it controls.
Anything that helps them advance in that direction must be fought fiercely.
Personally I don't think there's anything wrong with the fundamental concept of signed exchanges. The only problem is that it's just that: a signed exchange of content, which should have nothing to do with the domain name authority in the URL. By all means, display "Content from: a.com" in a box next to the URL, but don't change b.com to a.com in the URL as though it doesn't already have a well defined meaning.
The issue is that the technical meaning of the URL is very far from what most user think of.
Is the URL an address for NYT's server? Not really because you are actually hitting Fastly's server. So when NYT sets up a magical DNS config, it suddenly is fine, but using crypto to sign the package and serve it on a CDN that way, then it's suddenly "subverting the meaning of the URL"?
We can have a real discussion of what the meaning of a URL is, but I think your interpretation is unfair. I think it's entirely fair to argue that it makes sense for URLs to be an address to a specific content.
My argument is not really concerned with what most users think of, but humor me, what do they think of?
> So when NYT sets up a magical DNS config, it suddenly is fine, but using crypto to sign the package and serve it on a CDN that way, then it's suddenly "subverting the meaning of the URL"?
Yes, because HTTP/S scheme URLs have a definition that implies a meaning, which is subverted when you create exceptions to that meaning. NYT setting up a "magical" DNS config that resolves to some third party server is perfectly fine by that definition, and resolving one FQDN while displaying another is not. It's not sudden, this standard has existed in one form or another since 1994.
> We can have a real discussion of what the meaning of a URL is
Yeah, let's do that instead of harping on about what's fair and unfair. It's not a matter of fairness, it's a matter of standardized definitions. By all means, create a new "amp:" URI scheme where the naming authority refers to whoever signed the data and resolves to your favorite AMP cache, but don't call it http or https.
An example of where this occurs today is caching. You could be hitting a cache anywhere along the way. Hell you could be seeing an "offline" version, but the website would still show you the "address" of the content.
This is no different, you're hitting a different cache, but the "URL" you see is the canonical address of the content you are looking at, not where it was actually fetched from.
The only sense in which content is located anywhere is as data on a memory device somewhere. With the traditional URI in which the host part of the authority is an address of or a domain name pointing towards an actual host, you have a better indication of where the content is located than you do if this is misrepresented as being some other domain name which in fact does not at all refer to the location of the content.
The shift, if any, is that people may be less interested in where the content is located and more interested in its publishing origin.
> An example of where this occurs today is caching. You could be hitting a cache anywhere along the way. Hell you could be seeing an "offline" version, but the website would still show you the "address" of the content.
Yes, because that's how domain names work.
> This is no different, you're hitting a different cache, but the "URL" you see is the canonical address of the content you are looking at, not where it was actually fetched from.
It's different in the sense that a host name as displayed by the browser then has multiple, conflicting meanings that have no standardized precedent.
For your question about fastly, I already answered that in the comment you replied. The fastly CDN requires that the DNS is configured to point at fastly servers. Take a look at https://docs.fastly.com/en/guides/sign-up-and-create-your-fi... under "Start serving traffic through Fastly".
Once you’re ready, all you need to do to complete your service
setup and start serving traffic through Fastly is set your domain's
CNAME DNS record to point to Fastly. For more information, see
the instructions in our Adding CNAME records guide."
AMP requires that you consume other Google products, which requires that additional JS is loaded. When your mobile site doesn't use AMP, Google limits SEO rankings your mobile site can have. Google AMP requires your pages meet Google's Content Policies or they won't host them.
AMP and CDN delivered pages are architected differently and Google imposes restrictions and requirements that don't exist in a CDN.
I'm still opposed to the change, I see this centralization of the web through CDNs as a bad thing, I don't want to make it easier.
Mozilla (and Apple) are strictly against it and thank god for Mozilla. If Google had a bigger market share this would already be something we would have been living with. I'm sure there are better sources for this, but here is the first result:
https://9to5google.com/2019/04/18/apple-mozilla-google-amp-s...