5 ms·

As someone who has actually worked on end to end encryption at Meta, I can tell you I am not aware of anything where the company reads your WhatsApp messages -

As someone who has actually worked on end to end encryption at Meta, I can tell you I am not aware of anything where the company reads your WhatsApp messages - either in transit or device. The company takes fairly serious measures to ensure it cannot even accidentally infer such contents. I don't know what is happening in this specific case. Perhaps the ads came from some other similar search queries. Perhaps they came from the keyboard intercepting what was typed. Or perhaps something else that I can't think of. But I'm nearly certain it did not come from meta intercepting the contents of your messages. It's hard to convince people at this point because many have lost trust in Meta as a company, and I understand that. But I still find it stunning that so many people are making so many false claims without any actual knowledge to back it up.

donedealomg 4y ago
lrvick 4y ago
Google, Apple, or Meta retain the power to ship a tweaked binary with a compromised RNG to a subset of users if authorities order them to be it now or in the future after a privacy policy change. Proprietary encryption means users cannot verify or control the keys or the code that generates or uses the keys. The app can exfiltrate the keys or do any keyword processing on behalf of Meta as well which can include well intentioned features like forwarding paintext messages containing certain dangerous-seeming words to authorities or theoretically trusted third party review teams. Naturally they could also return -metrics- about frequency of word use back to Meta for ad targeting as well. I too have been a champion of encryption and privacy at past companies only to have all my work undone and watch all the data become plaintext and abused for marketing by a new acquirer. The only way end to end encryption solutions can avoid these types of abuses is when the client software is open source and can be audited, reproducibly built, and signed by any interested volunteers from the public for accountability. Short of that it is really not that much different than TLS with promises Meta will not peek, at least not directly, today.
- Melatonic 4y ago
  Even with end to end encryption couldnt the app at the end also be just aggregating the data (or even transcribing audio) to send over separately?
  - lrvick 4y ago
    Yeah though this is more likely to be detected with basic network analysis. Selectively compromising the RNG seed would be much harder to detect without source code.
- beiller 4y ago
  If they modified the RNG of person A's phone app during a forced stealth update, then shouldn't person B not be able to decrypt the message? Have you ever had an app update to Whatsapp that you cannot communicate with other people until you are forced to update? The alternative is that there is a vast internal conspiracy at meta that hundreds of engineers, and hundreds of ex-engineers are somehow silent on, which would be using 2 encryption keys, one that law enforcement can read, and one that the other end of the device can read. Isn't provable that Whatsapp the app is using the operating system level secure prng functions? If there was evidence of this, wouldn't it be great for a whistleblower to come out and make a killing shorting Meta's stock? Right now would be the perfect time to be kicked while they are down.
  - mhio 4y ago
    > then shouldn't person B not be able to decrypt the message? The RNG example is a way to create keys that make it trivial for "C in the middle" with the RNG details to extract the contents. They are still valid, just not useful as keys. The Juniper attack and Dual EC exploit is a good real world example of compromising an RNG for passive decryption, although Dual EC was designed to be like that. https://www.cs.utexas.edu/~hovav/dist/juniper.pdf https://www.cs.utexas.edu/~hovav/dist/juniper.pdf
pdntspa 4y ago
Are you absolutely sure that this is still the case? You say you "used" to work on it, but modus operandi for these companies is rugpulling protections like this as soon as nobody is looking
- xerxesaa 4y ago
  I feel quite confident based on first hand knowledge of code, system design, and the many, many privacy reviews we had to go through when building new features to ensure we didn't accidentally log or otherwise infer data we weren't supposed to. WhatsApp architecture is designed with the assumption that the server could be compromised and yet such an event should not result in any message contents being revealed. Furthermore, the encryption function is designed to ratchet and rotate keys so that a leak of a key at a given point in time would not compromise past and future messages. So yes, I have a strong sense of confidence that message contents are not exposed to Meta and, given the bar set by privacy reviews, I don't think Meta would do some backdoor workaround like scraping the contents off the device and sending an unencrypted copy. To be clear, my claims are specifically around message contents and when it comes to certain metadata (ex. the sender/receiver, the names of groups, etc) I don't recall the exact details of how they are treated. Now, despite the fact that I've said all this and that my knowledge on the matter is fairly recent, I'm not sure I could ever say anything with absolute confidence. The code base is huge and not open source. I obviously have not seen every line of code and as you pointed out, there's always a chance some company policy changes happened without my awareness. So I would say "highly" confident but not "absolutely" confident.
bartimus 4y ago
What about spell-check data?
hayst4ck 4y ago
> Perhaps the ads came from some other similar search queries. Perhaps they came from the keyboard intercepting what was typed. Or perhaps something else that I can't think of. But I'm nearly certain it did not come from meta intercepting the contents of your messages. Isn't this kind of splitting hairs? Does it matter if text information came from a "side channel"? It seems like the promise Facebook makes is that 'your communication using whats app is secure,' that's certainly my interpretation of what "end to end encrypted" means. It is a promise of security. That means text is sacred and even text sent to giphy should be privileged from the ad machine. The question being asked here is not "is it end to end encrypted?" It's "are my communications secure?" End to end encryption is just one element of that security.
- thetrb 4y ago
  The thing is if it's a 3rd-party Android keyboard or similar that logs your messages then there's nothing Meta can do about this.
  - hayst4ck 4y ago
    That's still Facebook's problem, no excuses. Facebook absolutely has the power and resources to lobby google and congress. Security teams at both companies will unequivocally agree that keylogging presents an extremely grave security risk that consumers are unlikely to understand the consequences of and therefore need to be protected from. Imagine a hapless military professional/politician downloading one. The problem is one of alignment. Facebook wants to monetize whatsapp and wants the whatsapp data. That's why there was a mass exodus to signal in the first place. Facebook was weakening the protections of the app. Due to the alignment problem Facebook can't advertise whatsapp as the secure and private choice because they are actively working to make it less secure and private. That's why Brian Acton quit (leaving $$$$$$$ behind) in the first place.
    - thetrb 4y ago
      That 3rd-party keyboard would also be able to log your Signal messages, so I don't get your point.
daqhris 4y ago
Thanks for your explanation. I didn't have in mind the scenario of a keyboard logging user inputs besides the normal functionality of WhatsApp. I find this theory to be very plausible. Not at all happy with Meta's privacy policy, but I agree that it is worth considering other threats. From using a VPN that logs all incoming and outgoing traffic (NetGuard) on an Android One device, I've noticied that the default Google keyboard gets in touch way too many times with some distant servers. Whereas, an open source keyboard from F-Droid, FlorisBoard, does no snooping and gets updated solely through the app store.
- AJ007 4y ago
  The third party keyboard apps are a big question for the OP. Another consideration, there are companies that track and sell geolocation data. It's "anonymized" but so precise you know the street address a user resides at. It is not a stretch to consider "anonymized" retargeting from keyboard inputs. I was dismissive of it in the past, as comments voted higher here are. However I've seen enough weird ads show up within minutes of making jokes about obscure topics that I suspect there is something going on. The piece that might be missing here is third parties collecting signals, "anonymizing" them, and then ads get re-displayed through Facebook, Google, etc. It may not be the major ad platforms doing it directly. In theory this should be harder now with the iOS tracking restrictions. For the skeptical, consider Avast's Jumpshot. Here millions of users thought they were protecting themselves when their raw browsing stream was being sold live to third parties. I They aren't the only company that has done that. https://www.theverge.com/2020/1/30/21115326/avast-jumpshot-subsidiary-suspended-data-collection-selling-ceo-blog-post https://www.theverge.com/2020/1/30/21115326/avast-jumpshot-s...
5d8767c68926 4y ago
Meta has repeatedly demonstrated they will do whatever it takes to capture user data. Kid VPNs, in app browsers, etc. Is it any surprise that people are deeply suspicious of any coincidences that arise from using a supposedly private channel.? Given evidence at hand, it is hard to view Meta as anything but a bad actor.

donedealomg 4y ago

lrvick 4y ago

Google, Apple, or Meta retain the power to ship a tweaked binary with a compromised RNG to a subset of users if authorities order them to be it now or in the future after a privacy policy change. Proprietary encryption means users cannot verify or control the keys or the code that generates or uses the keys. The app can exfiltrate the keys or do any keyword processing on behalf of Meta as well which can include well intentioned features like forwarding paintext messages containing certain dangerous-seeming words to authorities or theoretically trusted third party review teams. Naturally they could also return -metrics- about frequency of word use back to Meta for ad targeting as well. I too have been a champion of encryption and privacy at past companies only to have all my work undone and watch all the data become plaintext and abused for marketing by a new acquirer. The only way end to end encryption solutions can avoid these types of abuses is when the client software is open source and can be audited, reproducibly built, and signed by any interested volunteers from the public for accountability. Short of that it is really not that much different than TLS with promises Meta will not peek, at least not directly, today.

Melatonic 4y ago
Even with end to end encryption couldnt the app at the end also be just aggregating the data (or even transcribing audio) to send over separately?
- lrvick 4y ago
  Yeah though this is more likely to be detected with basic network analysis. Selectively compromising the RNG seed would be much harder to detect without source code.
beiller 4y ago
If they modified the RNG of person A's phone app during a forced stealth update, then shouldn't person B not be able to decrypt the message? Have you ever had an app update to Whatsapp that you cannot communicate with other people until you are forced to update? The alternative is that there is a vast internal conspiracy at meta that hundreds of engineers, and hundreds of ex-engineers are somehow silent on, which would be using 2 encryption keys, one that law enforcement can read, and one that the other end of the device can read. Isn't provable that Whatsapp the app is using the operating system level secure prng functions? If there was evidence of this, wouldn't it be great for a whistleblower to come out and make a killing shorting Meta's stock? Right now would be the perfect time to be kicked while they are down.
- mhio 4y ago
  > then shouldn't person B not be able to decrypt the message? The RNG example is a way to create keys that make it trivial for "C in the middle" with the RNG details to extract the contents. They are still valid, just not useful as keys. The Juniper attack and Dual EC exploit is a good real world example of compromising an RNG for passive decryption, although Dual EC was designed to be like that. https://www.cs.utexas.edu/~hovav/dist/juniper.pdf https://www.cs.utexas.edu/~hovav/dist/juniper.pdf

pdntspa 4y ago

Are you absolutely sure that this is still the case? You say you "used" to work on it, but modus operandi for these companies is rugpulling protections like this as soon as nobody is looking

xerxesaa 4y ago
I feel quite confident based on first hand knowledge of code, system design, and the many, many privacy reviews we had to go through when building new features to ensure we didn't accidentally log or otherwise infer data we weren't supposed to. WhatsApp architecture is designed with the assumption that the server could be compromised and yet such an event should not result in any message contents being revealed. Furthermore, the encryption function is designed to ratchet and rotate keys so that a leak of a key at a given point in time would not compromise past and future messages. So yes, I have a strong sense of confidence that message contents are not exposed to Meta and, given the bar set by privacy reviews, I don't think Meta would do some backdoor workaround like scraping the contents off the device and sending an unencrypted copy. To be clear, my claims are specifically around message contents and when it comes to certain metadata (ex. the sender/receiver, the names of groups, etc) I don't recall the exact details of how they are treated. Now, despite the fact that I've said all this and that my knowledge on the matter is fairly recent, I'm not sure I could ever say anything with absolute confidence. The code base is huge and not open source. I obviously have not seen every line of code and as you pointed out, there's always a chance some company policy changes happened without my awareness. So I would say "highly" confident but not "absolutely" confident.

bartimus 4y ago

What about spell-check data?

hayst4ck 4y ago

> Perhaps the ads came from some other similar search queries. Perhaps they came from the keyboard intercepting what was typed. Or perhaps something else that I can't think of. But I'm nearly certain it did not come from meta intercepting the contents of your messages. Isn't this kind of splitting hairs? Does it matter if text information came from a "side channel"? It seems like the promise Facebook makes is that 'your communication using whats app is secure,' that's certainly my interpretation of what "end to end encrypted" means. It is a promise of security. That means text is sacred and even text sent to giphy should be privileged from the ad machine. The question being asked here is not "is it end to end encrypted?" It's "are my communications secure?" End to end encryption is just one element of that security.

thetrb 4y ago
The thing is if it's a 3rd-party Android keyboard or similar that logs your messages then there's nothing Meta can do about this.
- hayst4ck 4y ago
  That's still Facebook's problem, no excuses. Facebook absolutely has the power and resources to lobby google and congress. Security teams at both companies will unequivocally agree that keylogging presents an extremely grave security risk that consumers are unlikely to understand the consequences of and therefore need to be protected from. Imagine a hapless military professional/politician downloading one. The problem is one of alignment. Facebook wants to monetize whatsapp and wants the whatsapp data. That's why there was a mass exodus to signal in the first place. Facebook was weakening the protections of the app. Due to the alignment problem Facebook can't advertise whatsapp as the secure and private choice because they are actively working to make it less secure and private. That's why Brian Acton quit (leaving $$$$$$$ behind) in the first place.
  - thetrb 4y ago
    That 3rd-party keyboard would also be able to log your Signal messages, so I don't get your point.

daqhris 4y ago

Thanks for your explanation. I didn't have in mind the scenario of a keyboard logging user inputs besides the normal functionality of WhatsApp. I find this theory to be very plausible. Not at all happy with Meta's privacy policy, but I agree that it is worth considering other threats. From using a VPN that logs all incoming and outgoing traffic (NetGuard) on an Android One device, I've noticied that the default Google keyboard gets in touch way too many times with some distant servers. Whereas, an open source keyboard from F-Droid, FlorisBoard, does no snooping and gets updated solely through the app store.

AJ007 4y ago
The third party keyboard apps are a big question for the OP. Another consideration, there are companies that track and sell geolocation data. It's "anonymized" but so precise you know the street address a user resides at. It is not a stretch to consider "anonymized" retargeting from keyboard inputs. I was dismissive of it in the past, as comments voted higher here are. However I've seen enough weird ads show up within minutes of making jokes about obscure topics that I suspect there is something going on. The piece that might be missing here is third parties collecting signals, "anonymizing" them, and then ads get re-displayed through Facebook, Google, etc. It may not be the major ad platforms doing it directly. In theory this should be harder now with the iOS tracking restrictions. For the skeptical, consider Avast's Jumpshot. Here millions of users thought they were protecting themselves when their raw browsing stream was being sold live to third parties. I They aren't the only company that has done that. https://www.theverge.com/2020/1/30/21115326/avast-jumpshot-subsidiary-suspended-data-collection-selling-ceo-blog-post https://www.theverge.com/2020/1/30/21115326/avast-jumpshot-s...

5d8767c68926 4y ago

Meta has repeatedly demonstrated they will do whatever it takes to capture user data. Kid VPNs, in app browsers, etc. Is it any surprise that people are deeply suspicious of any coincidences that arise from using a supposedly private channel.? Given evidence at hand, it is hard to view Meta as anything but a bad actor.