Startup unveils offline model to flag engineered and natural pathogens
Access stays limited to vetted testers while national labs and government agencies assess how well it performs.
Radical Numerics, a San Francisco startup founded in 2025 by researchers behind the Evo and Evo-2 DNA models, has unveiled Omnii Defense, a model meant to flag natural or AI-engineered pathogens. It runs on a laptop with a USB gene sequencer and works offline, which suits field researchers with unreliable connections. For each sequence it estimates the likelihood that the sequence is pathogenic or AI-generated, identifies the closest taxonomic group, and assigns a threat score from zero to one. Its appeal over older catalogue-matching tools is that it learns functional patterns rather than relying on exact sequence matches, and in a test where Bundibugyo virus was removed from training data it still flagged a third of those sequences as pathogenic. Access is limited to vetted testers while national laboratories and government agencies assess its performance, and outside experts caution that detecting a signal is far from confirming real danger or engineering.
Why it matters
The model tries to catch dangerous sequences, both natural and AI-engineered, even when the exact pathogen was absent from its training data. The test result is an early signal, not proof of performance. Outside experts note that a flagged signal is far from confirming whether a pathogen is genuinely dangerous or was engineered, so the tool's real-world value depends on evaluations still under way.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.