Hacker News Reader

Comment by Legogris | Hacker News Reader

Anecdata: I recently shared a private FB event link with a friend over WhatsApp (DM, not group chat) while we were in the same room. She saw and opened it. Some hours later it was gone from both of our chat histories on all devices, with no notice of deleted messages. Earlier and later messages were intact.

I don't see how this could happen if that claim was true.

View on HN

I have a more clear event: At my company we regularly communicate single use credit card numbers for company purchases. One of those was sent via WhatsApp, and stolen. Neither the origin phone nor the destination phone were compromised. The card was used to purchase Adwords, the transaction originated in the US (we are in Europe).

Our theory is that at least images on WhatsApp are human-reviewed, and one reviewer saw the credit card go through and took the opportunity.

We reversed the transaction and switched to Signal...

How do you know the card wasn't compromised before it reached you?

It's a virtual card, generated in the banking app, to be used one single time. The card never existed outside the bank IT systems or the phones involved in the communication.

> the bank IT systems

Eh ...

At least on android, most whatsapp data is stored in a folder named whatsapp, which can be accessed by any app with storage permission. So you can see all sent/received images and videos in photo viewer apps etc. Signal stores them in a place which at least in Android 11 can't be accessed by other apps.

Whatapp does have a list of banned hashes for images so that certain images cannot be send - this hash check is done on the client so as not to break e2e encryption.

The same can surely be done with web links.

Source?

> I don't see how this could happen if that claim was true.

I don't see the relation. They could (and I'm sure they do) attach a message id to the e2e-encrypted payload, shared between message sender and receivers. They could remotely delete messages by id, either by design or through a bug. None of this requires breaking e2e.

I mean, I'd argue that remote message deletion by id should not be possible (i.e. the client should not permit it) and certainly not without user notification, but that's a different matter.

Deleted message contained more text than only the link. The event was quite small and local (social event with <200 participants). The organizer (a friend of mine) had asked people to not share the link via social media.

While what you're saying is theoretically possible, I find it such a stretch that the only reasonable explanation save for some very unlikely bug is that message contents are indeed accessed in some form outside of our own devices.

It would be terrific to find a way to reproduce this.

What do you believe is the likely reason for the link to be deleted? Do you think that the organizer (your friend) did something at his end that caused the link to be deleted everywhere? I.e., he wanted to delete the link. In that case, it should be possible to reproduce this.

Or do you think Facebook or WhatsApp disapproved of the link and therefore deleted it? Was it something controversial or against Facebook rules? It could be possible to reproduce that as well if a group of users shares an equally controversial link.

I agree. It really beats me. The organizer did write on the event info page something like "don't share on SNS". While it's a curious coincidence that it's specifically this link that gets deleted, AFAIK there's no way for an organizer to prevent sharing apart from making the event private and requesting attendees in free-text to keep it to themselves.

The only thing that I think could potentially be controversial would be that it was a social gathering during the pandemic. The event page itself was and still is up. If it was deemed against FB rules I'd expect it to be reflected in some way on FB, and not just by deleting WhatsApp messages refering it.

I really can't give a good reason apart from "some ML model got triggered somehow, which prompted a human somewhere to look at it for a couple of seconds and click the delete button"

WhatsApp visits links to generate a preview, so presumably FB knows the content of your message because at least the link is leaked.

Yes indeed, to request the OpenGraph metadata. Though technically not needed I assume that any hyperlink will be readable by WA/FB. Anyone knows if these requests go via their own servers?

I just checked the web UI by pasting a link and didn't see a request go out to the domain, so I assume it went through the websocket to FB.

When did this happen? They hadn't implemented E2E Encryption before 2016.

During 2021.

Regardless of e2ee, they might have code bundled in the WhatsApp app that just deletes local video messages based on a hash. That wouldn’t break encryption.

It was an event link, think:

  > The event next week:
  > https://....

(Neither of us remember if those two lines were distinct messages or two lines in the same message; regardless, neither is there anymore)

Same message doesn't mean same hash. Usually a nonce is used and if not: WhatsApp uses the Signal Protocol, which uses different keys for each message

The client can delete the message if the decrypted hash matches a certain blacklist. It doesn't have to happen on the FB servers.

Such a mechanism allows governments to turn WhatsApp into a propaganda machine very easily, though, so I'm not sure if I would consider such a mechanism for my app if I were in a similar position.

You cannot and should not be able to use a hash or rainbow table for encrypted messages. If that is possible (you don't use a nonce), then your encryption is broken since many common messages can be looked up in a rainbow table and you can use replay attacks.

That would have to happen on the device before encryption or after decryption.

Whatapp's client applications already do this before encryption to check against a list of hashes of child abuse material.

This is of course great, but such hash based filters for images or videos can be circumvented easily by just flipping one bit (without corrupting the file). I guess what we really need is some kind of ML trained filter that has a great rate of success and a very low false positive rate.

Reply on news.ycombinator.com