Google DeepMind Embeds Watermarks in AI-Generated Proteins While Maintaining Biological Function

Google DeepMind has announced SynthID Bio, a watermarking technology that embeds imperceptible signatures directly into AI-designed protein sequences and predicted 3D structures—all while preserving their ability to function in living systems. The proof-of-concept demonstration, published in collaboration with Stanford University and the Arc Institute, addresses a growing biosecurity challenge: as AI models generate novel proteins at scale, tracking their provenance has become critical for preventing misuse and maintaining scientific integrity.