New 'Spymark' Technology Raises Privacy Concerns Beyond Traditional Watermarking

Defining 'Spymarks'

A new term, "spymark," is introduced to distinguish a type of hidden signal from traditional watermarks. While watermarks verify authenticity or ownership, spymarks are hidden signals that make digital work traceable without the user's knowledge or consent. This distinction aims to highlight the privacy implications of these newer technologies.

Google SynthID as an Example

Google's SynthID is presented as a spymark system that embeds secret signals, imperceptible to humans, into images, audio, text, and video. These signals can encode database identifiers linked to personal information such as full names, IP addresses, dates of birth, physical addresses, and political affiliations. The SynthID-Image paper reports that its SynthID-O variant can embed a 136-bit payload in a 512x512-pixel image, with 64 bits for a database identifier and 72 bits for error correction.

Widespread Development and Integration

SynthID is not the only spymark system under development; OpenAI and other tech companies are also creating similar systems. These companies state that spymarking can help identify AI-generated content, but the systems include robust tracking mechanisms beyond simple watermarking. It is anticipated that social media platforms, content production tools, and smartphones may soon incorporate spymarking algorithms to embed these signals into published content.

The Importance of New Terminology

The article argues for the new terminology to clarify the privacy risks associated with modern "watermarks." The term "watermark" has become too broad, encompassing both benign authenticity marks and concerning hidden tracking signals. By using "spymark," the aim is to bring the privacy concern to the forefront of discussions, differentiating clandestine surveillance and involuntary disclosure from traditional, consent-based marking.