Scientists watermark AI-designed proteins to aid provenance tracking.
They publishedScientists can now 'watermark' AI-generated proteins to deter bioweapon creation. But will it really work?
First reportLive Science · 23 Oct, 09:44Z
Latest reportLive Science · 23 Oct, 09:44Z
Span0 hours · two reports
Base — every candidate starts here5
Public-interest topic (politics, world, science, health…)+1
Public-interest terms in the text (courts, tariffs, ceasefire…)+1.25
High-reliability source boost+0.75
Significance8
Deterministic and identical for every story — no engagement signals, no personalization of the score.
Why it matters
What
Tests show marked proteins retain function, but experts say the identifiers can be removed.
Why
Removable marks limit their value as a standalone safeguard against misuse.
Watch
Whether developers adopt closed-source models.
Where the coverage comes from
Perspective mix: 1 center.
Lean and credibility are curated v1 mappings from public ratings (AllSides / MBFC); unknown outlets stay unrated, never guessed. How outlets are rated →
How it unfolded — every article
Summary
- 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.