Deep-TEMPEST: Recovering images from unintentional EM radiation of HDMI

Impressive! But I'll note to readers to not let the inclusion of HDMI imply anything special. There's not a lot that is different between HDMI and VGA or NTSC in the particular case of displaying bitonal images. I suspect running at higher resolution and displaying color would really throw off the ability to perform this intercept. I don't know where the ML is actually coming into it though; maybe helping to recreate sync or doing some kind of thresholding or phase correction on the scanlines maybe? In any case it doesn't look like anything new is happening here to my observation, but these twitter announcements are essentially useless for details, so I'll reserve judgement.

TEMPEST [1] Is a US Gov't program revolving around, essentially, emissions security. It has been known for decades that you can spy on someone using e.g. the EM emissions from their monitor/VGA cable. This is a tweet apparently highlighting a method of using deep learning to make recovered images easier to read. Not many details. This seems more related to image processing than to security, based on my first impressions.

[1] https://en.wikipedia.org/wiki/Tempest_%28codename%29?useskin...

I imagine using HDCP or DPCP would defeat this since encrypted data should look pretty much the same as noise. There is finally a use of these technologies beyond DRM.

Shades of Cryptonomicon, phreaking, and a flashy Perl script tweaking Finux to overcome the RF/Phreakers in the next room.