Spymarks, Not Watermarks

Spymarks are hidden, intentionally embedded signals in images, audio, text, or video that encode identifiers linking a file back to a user or account without that user’s meaningful knowledge or consent. Examples include Google’s SynthID family - SynthID-O can encode a 136‑bit payload into a 512×512 image (enough for a 64‑bit database ID plus 72 bits for error correction) - text steganography that steers word choices to represent bits, audio schemes that alter waveforms or spectral features, and even longstanding printer tracking dots that encode time, date, and serial numbers. Open-source tools such as audiowmark (from 2018) already hide 128‑bit payloads protected by AES keys, and many companies including major AI developers are building similar systems into content-creation and distribution pipelines.

The argument is that these hidden tracking marks are not benign watermarks or user-editable metadata like EXIF/ID3; they are covert, robust to compression and some edits, can survive metadata stripping, and are designed to map files to account records. That makes them powerful tools for tracing distribution, reconstructing social graphs, and exposing authors or distributors - posing clear risks to privacy, dissenters, and whistleblowers. Rebranding these mechanisms as “spymarks” forces attention on their surveillance intent and the urgent need to detect, resist, and limit their deployment.