This made me think of The Thing, a fascinating listening device made by the Soviets that was completely passive until powered remotely. Very difficult to detect when unpowered.
The Eye of Sauron paper does acknowledge the limitations that would likely prevent it from working against such a device. Impressive nonetheless. True privacy/security is just very hard!
the rabbit hole goes so much deeper than this 'Thing'.
An interesting read is Spy Catchers written by first science officer from MI5 in cold war. he was tasked with reversing listening devices and making them.. it doesnt spill all the beans but givea a good insight into how far they were already then.
u can hardly imagine what tech exists now 40-60 years later as electronics and computers have shrunk drastically and infinitely more weird material properties are discovered....
TSCM services commonly employ non linear junction detectors, if you have a legit need and money to spend and worry that your offices have been bugged (or are at risk of being bugged), you absolutely can hire them. This isn't really a new thing. More advanced highly portable RF equipment, portable spectrum analyzers and things built with relatively low-cost SDR that talk to Linux SBCs make the equipment to do this a lot less costly than it used to be.
If the intention is to RF "illuminate" an array of DRAM, portable GaN based amps in different bands capable of directing RF at an area (let's say, with a set of horn antennas on a plastic rolling cart being taken around an office) are also a lot smaller and less power hungry than they used to be.
Edit: Looking at the Zhang PDF, the box next to the laptop in figure 15 is a USRP 210 SDR, and a basic log periodic antenna on a PCB.
So really this boils down to "we're using a COTS SDR repurposed as a spectrum analyzer and we've written custom software to sniff what it looks like when DRAM is actively being used in a small embedded device". Some serious TSCM firms have been doing this for 35 years, just the equipment is a hell of a lot smaller and cheaper now.
Dan Gelbart's lecture on sensors[0] explains another fascinating method in detail using RF as well as the optical scanning method mentioned in the original video. The full lecture is worth a watch, but the timestamp is where the RF method explanation starts.
That Dan Gelbart lecture where he also explains how retroreflection[0] is used to detect remote viewing, was it anything with a lens or even human eye is really good.
Interesting - but 800-900MHz (the mentioned harmonic frequencies) are blasted with LTE and 5G in Europe.
But maybe with a sharp filter (there's a gap between up- and downlink) it can be used with a RTLSDR. But the RTLSDR hates close strong signals as its 8-Bit ADC doesn't have much dynamic range.
This is a genius approach. I was wondering how to approach this problem myself the other day, and other things like audio poisoning for ambient listening devices. This gives me hope for being able to actually fight in the cat and mouse game of privacy that's coming up.
Yes but only for detecting 5ghz WiFi cameras (actively streaming). If it’s 2.4ghz WiFi, or Ethernet wired, or it’s just recording to an sd card, it can’t be detected using the quadRF.
Right, but adding a big memory chip may be difficult in a video camera that must be as small as possible, except when using some expensive packaging method, with chip stacking.
Ensuring almost perfect shielding is cheaper and simpler. The camera should be completely enclosed in a metallic case, with only 2 small holes, for the camera lens and for the antenna output. Also its schematic should include adequate filtering components superposed to the integrated circuit packages, to minimize the length of the metal traces with variable potential, which radiate electromagnetic waves.
Improved shielding and filtering should reduce a lot the detection distance from the 20 meter range obtained with ordinary cameras, perhaps to a range under 1 meter.
If the camera is designed to not watch continuously, but to halt its activity for random intervals, it may have great chances to not be detected even by close sweeps with the camera detector.
They already use fiber-optic borescopes with no electronics on the lens side to create distance from electronics when their targets do sweeps.
You can also have opto-mechanical listening devices where the electronics are 100 m away. There is a vibrating membrane at one end, and a laser through the fiber reads the vibrations.
The speaker in the video was extremely difficult for me to understand but the idea seems quite clever.
What I'm most surprised about is just how well it actually works... I did not think such EM fluctuations would be usably detectable from several meters away... but I suppose this may be moreso due to improper/lack of shielding in the first place, which I would think should be easily fixable in most products to mitigate this type of detection.
Youtube subtitles can help in situations like this. Although they're not able to understand perfectly either. Example (1m0s) speaker means to say 'they can be disguised' but subtitles are:
> they can be disgusted
I think the speaker genuinely gets that word wrong; a malapropism (as opposed to the auto-generated English subtitles getting it wrong).
> The speaker in the video was extremely difficult for me to understand but the idea seems quite clever.
With a bit of practice it's actually quite easy to understand the accent of Chinese speakers as they have some typical patterns and speak slowly.
I must say, I loved listening to him because this was genuine human work. Human made slides (with lots of ugliness coming with PowerPoint), human typos, human grammar mistakes, all of it. It may be flawed in some ways but this fully removed any doubts that a slop machine came even near it. So refreshing in this day and age!
Shielding is addressed in the paper. They did some testing covering it with aluminum sheet (from coke can not aluminum foil) which decreased detection range from 30 to 2m. That's still not a bad range compared to other similar projects attempting to detect hidden cameras using EM. But with a properly designed shield the detection is probably even worse. Within limitations section they mention, they had problems detecting some smartphone cameras due them using low power memory and in general being better shielded.
Really interesting concept! I would not have thought of EMR variation as a usable side channel for this.
IIUC, the EMR increases following scene changes (like a light being turned on or off) are a result of the fact that video codecs try hard to compress similar images. So, I speculate that an "extremely bad codec" that simply dumps raw pixels to storage would not produce such variation, and therefore remain undetectable?
Makes sense. It might be possible to develop a codec that, in addition to being constant-time, has the property that every prefix of the bits encoding a unit (e.g., a frame or block within a frame) encodes an approximate version of that unit, where the longer the prefix the more faithful the approximation is. Then each unit could be output in full to storage, and the "output file pointer" could then be wound back to a point dependent on how many bits are actually required to represent it accurately -- so keeping the full amount for a full scene change, or just a few bits for the "same as last time" common case.
Oh, that is fascinating - I think a scheme like that could very well give you constant-space as well as constant-time, while calculating and preserving the desired compression information for use in transmission (which itself could separately be made constant for detectable channels).
This made me think of The Thing, a fascinating listening device made by the Soviets that was completely passive until powered remotely. Very difficult to detect when unpowered.
https://en.wikipedia.org/wiki/The_Thing_(listening_device)
And then I thought, something similar might be possible for video. Turns out, yep.
https://ieeexplore.ieee.org/document/8719264
The Eye of Sauron paper does acknowledge the limitations that would likely prevent it from working against such a device. Impressive nonetheless. True privacy/security is just very hard!
The Thing is full of surprises to this day.
> The Thing was designed by Soviet Russian inventor Leon Theremin, best known for his invention of the theremin, an electronic musical instrument.
the rabbit hole goes so much deeper than this 'Thing'.
An interesting read is Spy Catchers written by first science officer from MI5 in cold war. he was tasked with reversing listening devices and making them.. it doesnt spill all the beans but givea a good insight into how far they were already then.
u can hardly imagine what tech exists now 40-60 years later as electronics and computers have shrunk drastically and infinitely more weird material properties are discovered....
There's also the Moog synthesizer: https://www.google.com/search?q=moog+synthesizer
And the Moog company, specializing in high tech products, started by a cousin of the synthesizer guy: https://en.wikipedia.org/wiki/Moog_Inc.
Seems like the devices they found in the IBM Selectrics during Project GUNMAN (though, same ‘conflict’).
https://media.defense.gov/2021/Jul/13/2002761779/-1/-1/0/LEA...
Haseltine’s The Spy in Moscow Station is also a fascinating account of events.
TSCM services commonly employ non linear junction detectors, if you have a legit need and money to spend and worry that your offices have been bugged (or are at risk of being bugged), you absolutely can hire them. This isn't really a new thing. More advanced highly portable RF equipment, portable spectrum analyzers and things built with relatively low-cost SDR that talk to Linux SBCs make the equipment to do this a lot less costly than it used to be.
If the intention is to RF "illuminate" an array of DRAM, portable GaN based amps in different bands capable of directing RF at an area (let's say, with a set of horn antennas on a plastic rolling cart being taken around an office) are also a lot smaller and less power hungry than they used to be.
Edit: Looking at the Zhang PDF, the box next to the laptop in figure 15 is a USRP 210 SDR, and a basic log periodic antenna on a PCB.
https://www.google.com/search?client=firefox-b-d&q=USRP+B210...
https://www.ettus.com/all-products/ub210-kit/
So really this boils down to "we're using a COTS SDR repurposed as a spectrum analyzer and we've written custom software to sniff what it looks like when DRAM is actively being used in a small embedded device". Some serious TSCM firms have been doing this for 35 years, just the equipment is a hell of a lot smaller and cheaper now.
Dan Gelbart's lecture on sensors[0] explains another fascinating method in detail using RF as well as the optical scanning method mentioned in the original video. The full lecture is worth a watch, but the timestamp is where the RF method explanation starts.
[0] https://youtu.be/0MtRxX0crjU?t=1815
That Dan Gelbart lecture where he also explains how retroreflection[0] is used to detect remote viewing, was it anything with a lens or even human eye is really good.
[0] straight link to where he explains retroflection https://youtu.be/0MtRxX0crjU?t=3019
Also many other videos where he explains how to solve hard problems are really good too.
e: There seems to be commercial product using retroreflection advertised Youtube
https://www.youtube.com/watch?v=NaXSRpTeMbU
It’s really clever.
The idea is that cameras read/write to a disk. That cause electromagnetic radiation that is detectable.
Then this drone does [something/move/stimulus] to trigger more read/write/EM radiation.
But part of me can’t believe this is the cutting edge in 2026? Feels like 60s engineers would’ve thought about this
Intelligence services are very likely far ahead and for obvious reasons the public is not privy to it.
The final boss of detection doesn't need the camera to be powered at all.
https://en.wikipedia.org/wiki/Nonlinear_junction_detector
Interesting - but 800-900MHz (the mentioned harmonic frequencies) are blasted with LTE and 5G in Europe.
But maybe with a sharp filter (there's a gap between up- and downlink) it can be used with a RTLSDR. But the RTLSDR hates close strong signals as its 8-Bit ADC doesn't have much dynamic range.
This is a genius approach. I was wondering how to approach this problem myself the other day, and other things like audio poisoning for ambient listening devices. This gives me hope for being able to actually fight in the cat and mouse game of privacy that's coming up.
I'd have asked the speaker if he scanned his actual hotel room at the conference.
I'm actually quite certain that he did. Imagine the scoop, opening your conference presentation with what you found upstairs the evening before!
Something like this seems to be more practical? https://www.crowdsupply.com/scale-rf/quadrf
Yes but only for detecting 5ghz WiFi cameras (actively streaming). If it’s 2.4ghz WiFi, or Ethernet wired, or it’s just recording to an sd card, it can’t be detected using the quadRF.
So the mitigation is to have larger memory buffers so processing/emissions can be delayed (assuming simple shielding isn't enough).
Right, but adding a big memory chip may be difficult in a video camera that must be as small as possible, except when using some expensive packaging method, with chip stacking.
Ensuring almost perfect shielding is cheaper and simpler. The camera should be completely enclosed in a metallic case, with only 2 small holes, for the camera lens and for the antenna output. Also its schematic should include adequate filtering components superposed to the integrated circuit packages, to minimize the length of the metal traces with variable potential, which radiate electromagnetic waves.
Improved shielding and filtering should reduce a lot the detection distance from the 20 meter range obtained with ordinary cameras, perhaps to a range under 1 meter.
If the camera is designed to not watch continuously, but to halt its activity for random intervals, it may have great chances to not be detected even by close sweeps with the camera detector.
Eye of Sauron to counter Palantir. Makes sense.
Does it though? The Eye of Sauron represents the being so sufficiently single-minded that it corrupted the Palantir and weaponized them for evil ends.
AKA, you know, Peter Thiel.
This feels very impactful for national security and corporate espionage.
They already use fiber-optic borescopes with no electronics on the lens side to create distance from electronics when their targets do sweeps.
You can also have opto-mechanical listening devices where the electronics are 100 m away. There is a vibrating membrane at one end, and a laser through the fiber reads the vibrations.
The speaker in the video was extremely difficult for me to understand but the idea seems quite clever.
What I'm most surprised about is just how well it actually works... I did not think such EM fluctuations would be usably detectable from several meters away... but I suppose this may be moreso due to improper/lack of shielding in the first place, which I would think should be easily fixable in most products to mitigate this type of detection.
Youtube subtitles can help in situations like this. Although they're not able to understand perfectly either. Example (1m0s) speaker means to say 'they can be disguised' but subtitles are:
> they can be disgusted
I think the speaker genuinely gets that word wrong; a malapropism (as opposed to the auto-generated English subtitles getting it wrong).
> The speaker in the video was extremely difficult for me to understand but the idea seems quite clever.
With a bit of practice it's actually quite easy to understand the accent of Chinese speakers as they have some typical patterns and speak slowly.
I must say, I loved listening to him because this was genuine human work. Human made slides (with lots of ugliness coming with PowerPoint), human typos, human grammar mistakes, all of it. It may be flawed in some ways but this fully removed any doubts that a slop machine came even near it. So refreshing in this day and age!
Shielding is addressed in the paper. They did some testing covering it with aluminum sheet (from coke can not aluminum foil) which decreased detection range from 30 to 2m. That's still not a bad range compared to other similar projects attempting to detect hidden cameras using EM. But with a properly designed shield the detection is probably even worse. Within limitations section they mention, they had problems detecting some smartphone cameras due them using low power memory and in general being better shielded.
Okay who can build me one of these?
Back to microfilm for the spies
Really interesting concept! I would not have thought of EMR variation as a usable side channel for this.
IIUC, the EMR increases following scene changes (like a light being turned on or off) are a result of the fact that video codecs try hard to compress similar images. So, I speculate that an "extremely bad codec" that simply dumps raw pixels to storage would not produce such variation, and therefore remain undetectable?
Potentially, although raw pixel data presents its own challenges - one 1080p60 camera will fill a terabyte of storage in less than an hour.
A better mitigation might be to take a leaf from cryptography and develop a constant-time codec.
Like all I frames in H264 ?
Non-branching memory accesses?
Makes sense. It might be possible to develop a codec that, in addition to being constant-time, has the property that every prefix of the bits encoding a unit (e.g., a frame or block within a frame) encodes an approximate version of that unit, where the longer the prefix the more faithful the approximation is. Then each unit could be output in full to storage, and the "output file pointer" could then be wound back to a point dependent on how many bits are actually required to represent it accurately -- so keeping the full amount for a full scene change, or just a few bits for the "same as last time" common case.
Oh, that is fascinating - I think a scheme like that could very well give you constant-space as well as constant-time, while calculating and preserving the desired compression information for use in transmission (which itself could separately be made constant for detectable channels).
phenomenal
Tag 2024?