Google DeepMind proposes watermarking AI-designed proteins

Google DeepMind published a research paper on Wednesday proposing a protein watermarking system. The method is designed to identify AI-designed protein sequences without compromising protein function, according to the paper.

The watermarking approach is based on Google's SynthID technology, which embeds signals into AI-generated content. In this case, the signal is embedded into protein sequences themselves.

The proposal addresses a gap in biosecurity screening. Nearly a year ago, a risk was flagged that software used to identify DNA sequences encoding potentially threatening proteins does not pick out AI-designed proteins.

ProteinMPNN, a tool for protein design, was developed by the Baker Lab. David Baker was honored with the same Nobel Prize shared with the head of DeepMind.

Ars Technica reported on the research paper. The paper does not announce a product release or deployment timeline.