Radical Numerics: defense losing bio-security race

Radical Numerics co-founder Eric Nguyen's path into this work runs through Stanford, where he couldn't get traction on Genomic Language Models for a long time because biologists didn't believe it would work, didn't think they could verify the output, and didn't see applications beyond what they could already do. He kept pushing, helped lead the development of Evo and contributed to Evo 2 at Arc Institute. Those models were later used by a separate Arc/Stanford team to generate entire bacteriophage genomes that were synthesized into functional viruses — an early sign of what these models can do.

The company's argument sits alongside a broader debate about open defensive tools. Clem Delangue argues cyber-warfare defensive capabilities need to be open and keep pace with frontier models' attack capabilities, and Anthropic's filters flag two areas, cyber-security and biology. Radical Numerics extends that logic to biology: genomic language models already handle RNA and protein because DNA sequences carry clear markers for genes, and the company tested whether a model shown progressively better aptamer scores could continue the trajectory on its own — it recapitulated some higher scores it had not been shown, which the company calls chain-of-thought in DNA. The stakes, as the company frames it, are whether pushing the frontier harder keeps defense from falling further behind.