An AI Wrote Sixteen Viruses That Never Existed, and the System Meant to Catch Dangerous DNA Was Built to Recognize Nature
About This Episode
Researchers at Stanford and the Arc Institute reported in Science on Thursday that generative AI models wrote complete bacteriophage genomes from scratch, and 16 of them worked — infecting and killing antibiotic-resistant E. coli. A companion editorial by Johns Hopkins biosecurity specialists published alongside it says the governance needed to steer this capability does not yet exist, because DNA synthesis screening is voluntary and matches orders against sequences nature already produced.
Our Take
AI stopped editing life and started writing it from scratch — and the checkpoint meant to catch a dangerous DNA order works by recognizing nature, which is the one thing a machine-written genome is not.
Continue Reading on Unscarcity
Human-in-the-Loop: Where AI Agents Must Stop
The Evo models wrote each genome in a single uninterrupted pass with no human edit inserted mid-sequence, which makes this the exact design question the article poses: not whether to constrain an autonomous system, but where a human has to stand in the path.
Anthropic Pulled the AI Alarm. Who Can Act on It?
Johns Hopkins biosecurity researchers used the same journal issue to say the oversight machinery for AI-written genomes does not exist — the article's core point that capability arrives before anyone has built the brakes.