Startup Launches AI Tool Designed to Detect Engineered Bioweapons

Frontier artificial intelligence laboratories have frequently warned that advanced systems could aid in the creation of dangerous biological agents. On Wednesday, startup Radical Numerics introduced a defensive countermeasure named Omnii Defense. The edge model operates locally on a laptop linked to a USB gene sequencer without requiring an internet connection, making it accessible for researchers working in remote field conditions.

Eric Nguyen, co-founder and chief executive of Radical Numerics alongside former Stanford doctoral researchers, established the San Francisco-based company in 2025. The firm previously utilized its underlying genome language model, Omnii, to develop custom bacteriophages that targeted a drug-resistant strain of E. coli under special FDA authorization. Their latest software evaluates genetic sequences to predict pathogenicity, whether a sequence is AI-generated, and assigns a unified threat score ranging from zero to one.

Traditional biodefense tools rely primarily on cataloged matches against known organisms, whereas machine learning models can recognize functional patterns even when genetic spellings are altered. In company tests where the Bundibugyo virus was scrubbed from training data, the AI successfully flagged significant portions of the viral sequences as pathogenic, while conventional tools showed negligible risk indication.

Access to Omnii Defense is currently restricted to vetted external testers, with the company collaborating alongside government agencies, national laboratories, and independent evaluators to examine the model's accuracy. Experts note that while signal detection represents progress, determining whether a specific pathogen poses real-world consequences remains a complex challenge.