Researchers developed a method to watermark proteins designed by artificial intelligence, enabling identification of AI-generated molecules.

Nature, Ars Technica Science, Science (AAAS) News · Science & Medicine

1 Oct 20263 outlets

What it means

DeepMind's embeds a secret, function-preserving into protein sequences produced by AI design tools, so labs can tell that a protein was AI-made even after the sequence has been mutated or redesigned. The goal is traceability: deterring bioweapons misuse and protecting scientific credit, while keeping the designed proteins' structure and activity intact. The new Nature paper demonstrates the watermark works in real lab experiments, with the caveat that a determined attacker could still strip it.

Accuracy & sourcesSourced

The claim is a fair summary of peer-reviewed research by Google DeepMind published in Nature, and it is reported by Nature's own news desk alongside independent coverage in Science, so the reporting sits close to the primary source. The main nuance is that this is a proof-of-concept: the watermark survives normal redesign pipelines but is not designed to resist a skilled adversary deliberately trying to remove it.

How new

Genuinely new: the underlying paper and the wave of news coverage about it both appeared within the past day, with no recirculation year given.

Researchers developed a method to watermark proteins designed by artificial intelligence, enabling identification of AI-generated molecules.

What it means

Accuracy & sourcesSourced

How new

Go deeper