Follow

With the world in lockdown, most "white-collar" crime (AKA "world-destroying corruption and looting") now takes place over Zoom. If you witness such a crime, you might be tempted to record the meeting and leak it to a journalist.

But leaking Zoom recordings is seriously fraught because they are full of personally identifying details. Some of these are "traitor-tracing" mechanisms, others are intrinsic to Zoom, and still others come from your end of the recording.

theintercept.com/2021/01/18/le

1/

· · Web · 2 · 14 · 9

Nikita Mazurov's guide to anonymizing Zoom videos for The Intercept tackles each of these classes of identifiers.

Traitor-tracing: Zoom meeting hosts have the option to add visual and audio watermarks to their videos.

The visual ones are perceptible - displaying your name/email on the screen so that it will be present in any video-grab - but the audio watermarks are a series of ultrasound chirps with unique identifiers in them.

2/

It's not clear where the audio watermark is inserted; Marzurov hypothesizes that the ultrasonic watermarks are inserted by your copy of the Zoom client, so using an external recording tool might bypass them.

Another important identifier in Zoom recordings is the arrangement of the other participants; this is different for every viewer.

3/

Any recording will reveal information that could identify the leaker: not just the user's OS, but also pop-up alerts about emails and messages.

Source protection with Zoom captures is really hard - but that's all the more reason for this discussion to take place in earnest.

eof/

@pluralistic Thanks for this relay. We should proselytize for Jitsi Meet.

@pluralistic
Interesting, an ultra-sound audio signature can trivially be wiped out by compressing down to 4k sound quality. But the amount of data to transmit is small enough that it can be sent using a low frequency wave, either a series of low volume thuds and thumps (AM) or by time-shifting the audio of the other speakers (FM). Getting rid of this without destroying the audio would be impossible without knowing how it's done, and the "how it's done" could constitute a per-call secret key.

@cjd @pluralistic

For AM, would it help if low and low-mid frequencies were filtered out of the audio, since we don't really need those to understand human speech?

Also, for FM, if there was a randomised time-shift pattern applied over the audio prior to leaking it, would that ruin the embedded signature?

@RazyDazzle @pluralistic
My instinct is that if they choose to pursue traitor-tracing in a serious way, it becomes a cat-and-mouse game which will sometimes be won and sometimes be lost...

@cjd @pluralistic

Yes, true, it will turn into an arms race, and the uncertainty alone will have a chilling effect on potential leakers.

Sign in to participate in the conversation
La Quadrature du Net - Mastodon - Media Fédéré

The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!