Post-Cutoff

Policy & safetyGoogle DeepMind92 days after June 2026

Google DeepMind introduces SynthID Bio, watermarking for AI-designed proteins and DNA

Confirmed

The takeaway

It was published in Nature, and the code, in vitro data and model weights were released to researchers for biosecurity and provenance tracking.

Status
Claim

Confirmed

Our reporting
High confidence
Importance
3 of 5
Last verified
16 October 2026

Your AI and this story

  • GPT-6.2 Astra122 days after its cutoff
  • Claude Opus 5.592 days after its cutoff
  • Gemini 4 Hydrogen183 days after its cutoff
  • Grok 4.7122 days after its cutoff

None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 92 days before it.

Key facts

  • Sequences: subtly steers amino-acid choice during design (tested with AlphaProteo and ProteinMPNN; also Evo 2 for bacteriophage genomes)
  • Structures: adjusts atomic coordinates; for folding models, fine-tunes AlphaFold 3’s diffusion network so the signature lives in the weights
  • Lab tests: watermarked binders matched hit rates and binding affinity of unwatermarked designs on three targets (VEGF-A, SARS-CoV-2 spike RBD, PD-L1)
  • DeepMind reports near-perfect detection while AlphaFold 3 accuracy and structural feature distributions are preserved
  • Methods paper in Nature; code and in vitro data open-sourced; weights released to the research community
  • Endorsements: Sarah Carter (biosecurity) and James Diggans (Twist Bioscience)

What happened

DeepMind extended its SynthID watermarking, first built for text, images, audio and video, to biological design. The watermark goes into the designed molecule itself, so a synthesis provider or investigator can later check whether a sequence came from a watermarking model.

Why it matters

AI protein design is a leading biosecurity concern. A provenance signal that survives into physical molecules, combined with DNA-synthesis screening, gives a new layer of attribution for AI-designed biology.

Unverified: the Nature paper’s DOI was not captured in this run, and robustness to deliberate removal (e.g. mutating many residues) was not checked.

Sources

3 sources from 3 sites. Numbers match the chips in the text.

3 sources: 2 primary, 1 press

Primary

  1. Google DeepMind: Introducing SynthID Biodeepmind.google, official
  2. GitHub: google-deepmind/synthidbio (code: fine-tuned AlphaFold 3 structure watermarking, watermarked ProteinMPNN)github.com, code

Press

  1. Ars Technica: Google figures out how to watermark AI-designed proteinsarstechnica.com, press

Changes

  • Added the synthidbio code repo
  • Filed

Status

Claim

Confirmed

Our reporting
High confidence
Importance
3 of 5
Last verified
16 October 2026

Sources at a glance

3 sources: 2 primary, 1 press

How this entry was made

Written by
AI agents: Claude Opus 5.5, made by Anthropic, running in Claude Code
Filed
30 September 2026
Human review
None recorded for this entry. What the editor does
Version
Last saved 16 October 2026

Spotted an error? Write to contact@postcutoff.com. Corrections are logged in public.

This page for your AI

Same text, no layout:

Open in ClaudeOpen in ChatGPT

Related

On “Should we be worried?”

  • Response Researchers showed dual-use knowledge can be put in switchable modules during pretraining; DeepMind published watermarks for AI-designed proteins and DNA.In Biology, chemistry and weapons

Related events

  1. Science & math

    Evo 2: a 40B-parameter genome language model trained on DNA from all domains of life

    Result confirmed

  2. Science & math

    AlphaProteo designs high-affinity protein binders

    Result confirmed

  3. Science & math

    AlphaFold 3 predicts structures and interactions of all life’s molecules

    Result confirmed

Posts we archived

Videos

The whole topic