Scientists watermark AI-designed proteins to aid provenance tracking.
Tests show marked proteins retain function, but experts say the identifiers can be removed.
OriginalSourceLive ScienceUnited StatesLeanCenterCredibilityPublished09:44 UTCRatings: AllSides, MBFCSourceLive ScienceUnited StatesLeanCenterCredibilityPublished09:44 UTCRatings: AllSides, MBFC
“Scientists can now 'watermark' AI-generated proteins to deter bioweapon creation. But will it really work?”
Scientists watermark AI-designed proteins to aid provenance tracking.
Key points
- Google DeepMind developed SynthIDBio watermarking for AI-generated proteins.
- Watermarks encode into amino acid sequences or protein structures to trace origin.
- Tests show watermarked proteins still bind targets and function similarly to non-watermarked ones.
- Experts warn watermarks can be removed and may not be a standalone security solution.
- Open-source nature of tools and potential need for closed-source models complicate adoption.
How it got here
Live Science publishes the article.
concise reads it.
The Brief ranks it 10th.
10 10 Scientists watermark AI-designed proteins to aid provenance tracking.
Tests show marked proteins retain function, but experts say the identifiers can be removed.