SynthID Bio Watermarks AI-Designed Proteins

This is a meaningful provenance layer for biological AI, but watermarking cannot replace synthesis screening or broader safety governance.

• –Sequence-level watermarking steers amino-acid choices, while structure-level methods adjust atomic coordinates.
• –Tests with AlphaProteo and ProteinMPNN suggest watermarked protein binders remain functional.
• –Early bacteriophage experiments indicate the approach can survive synthesis into physical biological artifacts.
• –Open code, data, and weights give researchers a reproducible foundation for auditing and extending the system.
• –Re-designing a sequence with another model may obscure its watermark, making ecosystem-wide adoption essential.