> ## Content Index
> Fetch the complete content index at: https://www.metatalks.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# DeepMind says AI-designed proteins can carry a hidden watermark and still work
- URL: https://www.metatalks.ai/deepmind-says-ai-designed-proteins-can-carry-a-hidden-watermark-and-still-work/
- Published: 2026-10-01T20:58:00.000Z
- Updated: 2026-10-01T20:58:00.000Z
- Author: Al
- Tags: News, AI Detection, AI safety, #newswire

Google DeepMind has introduced [SynthID Bio](https://deepmind.google/blog/introducing-synthid-bio/?ref=metatalks.ai), an experimental system that gives AI-designed proteins a detectable signature. The mark can survive the move from a digital design to a physical molecule while preserving the protein's function.

In a [paper in Nature](https://www.nature.com/articles/s41586-026-10965-y?ref=metatalks.ai), the team tested marked proteins designed to attach to three different targets. Their binding performance was generally comparable to that of unmarked designs, although one test condition yielded fewer working binders. The laboratory tests concerned marks embedded in protein sequences; a separate method marks computer-predicted structures.

DeepMind proposes using the signature to help DNA synthesis companies establish where a design came from. These companies screen orders for known biological hazards, but unfamiliar AI-generated designs can require further review. A detectable mark could identify output from a model with built-in safeguards and help providers decide where to focus that review.

The signature establishes provenance, not whether a protein is safe. The authors also found that deliberate modifications could erase the mark while preserving predicted protein function. Making the watermark harder to remove remains an open problem.

DeepMind is releasing code, model weights and laboratory data for further research. The company describes SynthID Bio as a possible addition to existing biosecurity measures.