[@BennJordan] There's a ZIP Archive Stored In This Video

Link: https://youtu.be/fhoJK02oD_E

Duration: 21 min

Transcript: Download plain text

Short Summary

A video demonstration of four modern covert data techniques — malware in Facebook photos, encrypted audio in Spotify tracks, ultrasound messages, and ebooks inside Instagram videos — paired with a survey of steganography history from 499 BC to Cold War KGB microdots. The creator, who built JABIT steganography tools and a patented AI music detector based on DCT compression forensics, explains compression tradeoffs, demonstrates hiding an SMB exploit inside a photo of their dog Gary on Facebook, and proposes a "pyramid of expectations" for audio embedding. Co-presenter Ben adds acoustic-modem work, including a fork of GG Wave called Wavist and his own 16-FSK "bum16" ultrasonic modem.

Key Quotes

1. "military historians estimate that cracking Enigma shortened the length of World War II by nearly half a decade, thus saving millions of lives and possibly changing the political map of Europe forever." (00:05:21)
2. "So, for many years, the Romans sent piss messages back and forth to and from spies." (00:02:37)
3. "All of these three things are possible, but one of those will need to be sacrificed in order to have the other two." (00:13:56)
4. "Basically, human brains are really good at noting subtle differences in smaller quantities and really shitty at noticing them in larger quantities." (00:10:37)
5. "So if we combined our most functionally robust audio and video decoders together, we'd be lucky to get like 100 kilobytes per second." (00:19:57)

Detailed Summary

Episode Overview

The episode showcases four modern covert data techniques and walks through the history, math, and hands-on results of steganography.

Four Demonstrated Techniques

• Hide encrypted files inside normal-sounding music via a web app that unlocks a secret audio stream from a Spotify link.
• Send encrypted messages via ultrasound using a phone as a modem, with no cell service or Wi-Fi.
• Exploit Meta's copyright protection to store ebooks inside Instagram videos.
• Store, host, and share malware inside a Facebook photo.

Historical Timeline of Steganography

• 499 BC — Greek ruler Histiaeus allegedly tattooed a message on a shaved servant's scalp, waited for hair regrowth, and sent him into Persian territory; this is presented as possibly the first recorded steganography and is said to have instigated the Ionian Revolt and Greco-Persian War.
• Greco-Persian Wars — Exiled Greek Demaratus scraped wax off a tablet, carved news of a Persian invasion plot into the wood, and reapplied the wax; Queen Gorgo of Sparta revealed the hidden message.
• 1499 — German monk Johannes Trithemius wrote Steganographia, which was banned by the Catholic Church; the books were recognized as cover texts for cryptography in 1606, and the third book was deciphered in 1996 after roughly 400 years.
• 1550 — Gerolamo Cardano published The Subtlety of Things, introducing the Cardan grill: a physical decryption key placed over text to reveal a hidden message.
• Enigma machine — Originally Dutch for banking communications; Germany bought the patent in 1923, then militarized it with plugboards, rotors, and a reflector so each keypress encrypted differently.
• WWII impact — Military historians estimate cracking Enigma shortened World War II by nearly half a decade, saving millions of lives.
• Cold War — KGB microdots, tiny film ciphers smaller than a grain of sand, were concealed in objects like a US quarter; the host shows a replica hidden inside a wedding ring that opens when slammed against a desk.
• Number stations — Shortwave transmitters still broadcast mysterious codes/text in Morse via ionosphere bounces, though less frequently than during the Cold War.

Compression Theory (DCT)

• Discrete cosine transform is presented as the core compression technique that keeps videos from being 200 GB and lets 5 minutes of Instagram browsing stay under a data cap.
• A 33-megapixel DSLR captures electrical charge for over 32 million pixels per photo, each storing color data.
• In an 8x8 grayscale block, DCT produces 64 coefficients, but the creator showed this can be reduced to 24 before meaningful visible change appears — about two-thirds less data.
• 8-bit images store each RGB channel as 0–255, yielding 16.7 million possible color combinations covering nearly all visible-light colors.

Image Steganography

• Most people don't notice color changes until the value shifts by about 4 or 5, creating an opening for least-significant-bit (LSB) steganography.
• JABIT Steganography hides files in JPEG lower frequencies the compression normally ignores, then brightens the affected pixel group; the creator built a web interface on top of the original source code to support hiding archives and non-text files.
• The creator successfully hid a Nexpose exploit for Microsoft Server Message Block (SMB) inside an image of their dog Gary, uploaded it to Facebook, downloaded it, and recovered the exploit — described as a major dangerous exploit surviving Facebook's image processing.

Audio Steganography ("Pyramid of Expectations")

• The creator frames audio steganography as a tradeoff between preserving the original sound, storing a meaningful payload, and surviving re-encoding by streaming services — at most two of three can be achieved at once.
• Stereo trick: make the file mono, invert one channel's waveform, and paste the hidden audio at a lower volume; in mono playback only the inverted difference (the transmission) remains.
• Spectral modulation inserts random chunks of inaudible noise in the audio spectrum, operating opposite to DCT.
• A test track (soft analog synth melody, no drums) ripped from Apple Music and Tidal at highest quality sounded identical to the original; the hidden layer carried science jokes. At 360p/480p YouTube compression, the hidden layer was audible but intelligible words were lost.
• LSB in the spectrogram: a 344 KB zipped ebook hid inside an 86-second recording, producing output indistinguishable from the original and supporting any file under 400 KB.
• The LSB embedding survived Spotify and Tidal high-quality settings but was stripped or broken by Apple's standard AAC compression and YouTube.
• For robustness on lossy platforms, multi-frequency shift keying was added, achieving about 2 KB per minute of audio at the cost of being more noticeable.

Acoustic / Ultrasonic Modems

• Georgi Gerganoff created GG Wave, an acoustic messaging protocol that ran on Android.
• Ben forked GG Wave into Wavist with updated libraries, adding audible/ultrasound/inaudible ranges, call signs, and encryption.
• Test scenarios included Bluetooth speakers, embedding messages in media, and hacking an FM radio signal: using an SDR to receive/decode FM, adding ultrasonic messaging noise, then re-encoding and transmitting on the same frequency to a car stereo.
• bum16 ("Ben's Ultrasonic Modem") combines encrypted communications with steganography, uses 16-FSK by default, supports parallel shift keying for quiet environments and forward error correction for noisy ones, and is free/open source with a rough Android build.

AI Music Detection Side Project

• The creator states they created and patented an AI music detection algorithm that forensically detects fragments of DCT streaming compression in training data, because AI music sites typically rip from Spotify/YouTube rather than obtaining master files (e.g., Pro Tools, FL Studio) from artists.