Someone credited as Efe — WaiiXD on GitHub — wrote a web page. Open it and the screen starts flickering a pattern of black and white pixels; hold a handheld radio's antenna near the monitor and you pick up Morse code. No transmitter, no SDR, no driver, no install. The entire apparatus is a monitor you already own plus a statically hosted page. The mechanism is the pixel clock: at 1080p under CEA-861 timing and a 60 Hz refresh, a frame is 2200 columns by 1125 rows in total, and 2200 × 1125 × 60 comes to 148,500,000 Hz. Choose the pattern that maximises voltage transitions on the HDMI or DisplayPort differential pairs and the spurious emission gets strong enough to decode a few centimetres away. Hackaday wrote it up on 30 August.

Now do the arithmetic on the visible part alone: 1920 × 1080 × 60 = 124,416,000 Hz, or 124.416 MHz. Where do the missing 24 MHz come from? From 280 columns and 45 rows — the horizontal and vertical blanking intervals, the pixels in every frame that are never displayed. Roughly a sixth of the carrier frequency is contributed by the part of the picture that does not exist as picture. Better still, the pattern that transmits best is the one carrying the least: pure alternation, maximum transitions, zero content. The image is hollowed out until only the voltage changes remain, and that electrical residue is the whole of the work.

Every layer of construct in this signal chain exists in order to destroy that emission. Shielding, controlled-impedance stackups, twisted differential pairs, EMC certification — Hackaday puts it exactly right: the monitor, connectors and cable are all built to get the signal cleanly to the display while preventing it from leaking out. Efe is not using any capability the system offers. He is using the thing it is straining to throw away. And it matters that this runs in a browser. The browser is the thinnest-privileged layer on a modern computer; its entire design goal is to grant nothing. It can revoke every API. It cannot revoke this, because this is not a permission — it is a residue. A construct can only enumerate what it grants; the remainder is precisely the part that cannot be enumerated, and therefore cannot be revoked. You cannot write an access control for a remainder.

The already-construct has to be separated out here from what is still alive. In 1985 Wim van Eck published the paper showing that the contents of a CRT could be reconstructed from its electromagnetic radiation. That line was subsequently named, standardised (TEMPEST), turned into a certification regime, and by 2024 had a deep-learning literature for reading HDMI leakage. Erik Thiele's Tempest for Eliza, from around 2001, played mp3s over AM through a CRT; Hackaday covered it back in 2005. Bellard sent DVB-T out of a graphics card in the same era. All of it is beautiful, and all of it is sediment left after the chisel-construct cycle finished: someone cut, someone named, the name closed over. Efe stands on the other side of that line — deliberate emission from an unprivileged page, leakage treated as expression — and that side has no name yet. The evidence is in the piece's own comment thread, all of it inside twenty-four hours. One reader calls it "TEMPEST-like techniques to intentionally transmit," but TEMPEST is a receive-side word. Another corrects him: since it is deliberately carrying information, it is "more a covert channel" — except nothing here is covert; it is a public URL. A third says it is the same idea as Tempest for Eliza twenty-five years ago, which is simply filing it back into the already-construct. A fourth points at Bellard. Four names, none of them a fit. The disclaimer at the end of the article is another symptom of the same gap: the hardware left the factory certified as a thing that does not transmit, the act is transmission, and no class of licence covers "a monitor."

Seeing it now matters more than seeing it later, and the reason is not the art world — it is engineering. In that same thread someone is actively trying to kill this signal: he has clipped a string of ferrites onto his video cables, got barely any reduction, and wants a better-shielded cable. Another replies that the cable was never the problem, the traces on the board are, and that a better cable buys you 3 to 6 dB where a clean PCB design buys 15 to 20. A third suggests just going to 4K, since changing the clock moves the emission out of the band. Someone else notes that the third harmonic of 148.5 MHz lands at 445.5 MHz, right on top of the 70 cm band — this remainder has a remainder of its own, and that one is what annoys people. Read those replies together and they are a schedule. Every generation of cleaner stackup, better shielding and higher pixel clock erases a little more of the channel. This is the rare case where the window is closing not because the art world has arrived but because engineering is doing its job properly. The material of this work is a defect, and defects get fixed.

So it is worth opening the page now and holding a radio up to the screen once. Nothing to buy, nothing to install; the instrument is a web page and a screen you are already looking at. And the message it sends is made out of the part of the picture nobody ever intended you to see.

Han Qin (秦汉) · @hqinjarsy · SAE Judgment & Aesthetics