Spymarks, Not Watermarks

Brandon Thomas proposes a new word for a troubling evolution of the "watermark" — the spymark. Where a watermark is a visible mark that verifies authenticity or asserts ownership, a spymark is a hidden signal that makes your work traceable without your knowledge or consent.

How it works

The headline example is Google's SynthID, which embeds signals — "imperceptible to humans," in Google's own words — into images, audio, text and video. Such a signal can encode a database identifier mapped to your identity: user records, full name, IP addresses, date of birth, physical address, even political party affiliation. Google reports that its SynthID-O variant packs a 136-bit payload into a 512×512-pixel image — enough for a 64-bit identifier with 72 bits left over for error correction.

SynthID was neither the first nor the only such system. OpenAI and others are building similar mechanisms. Officially they are pitched as a way to identify AI-generated content, yet the built-in tracking goes well beyond a simple watermark.

Why a new word

"Watermark" has become a catchall — banknote security, copyright overlays, hidden tracking signals alike. That last use obscures the privacy risk. Invoking Ursula K. Le Guin ("to speak the name is to control the thing"), the author splits the concepts apart. Spy- denotes clandestine surveillance and involuntary disclosure; -mark the embedded signal. The privacy concern is thus baked straight into the term.

Marks everywhere

Audio spymarks work on the same principles and are usually inaudible; they survive compression and re-encoding. The open-source tool audiowmark appeared back in 2018 and hides 128-bit payloads under an AES key. In text, SynthID steers word choices to form a statistical pattern that encodes a payload. The author also cites printer tracking dots, known since the 1980s, as an early spymark.

How spymarks differ from metadata

Standard metadata — EXIF in photos, ID3 in MP3s — is documented and can be inspected, edited and stripped. A spymark embedded in pixels, audio or word choices is invisible and often survives even metadata removal and some edits.

The danger is sharpest for whistleblowers and anyone who does not wish to be persecuted for their words or affiliations. The author sketches a future in which every device is attested and every post carries an account-linked spymark — halfway, he says, to Orwell's 1984. The remedy is simple: call a spymark what it is — a tool used to spy on you.

Source: Hacker News