A Startup Wants to Fight AI Bioweapons With AI

A San Francisco-based startup, Radical Numerics, has unveiled an innovative defense tool against AI-engineered pathogens. The company's new edge model, Omnii Defense, is designed to identify both naturally occurring and AI-generated pathogens using a laptop and a USB gene sequencer. Impressively, it can operate without an internet connection, making it ideal for researchers in the field with spotty connectivity.

While some experts argue that the threat of AI-made bioweapons is minimal, Radical Numerics believes its technology could prove invaluable in helping researchers and public health officials pinpoint emerging biological threats.

Eric Nguyen, the cofounder and CEO of Radical Numerics, anticipates a biodefense arms race akin to the current cybersecurity landscape, where threat actors and defenders are both leveraging AI to gain a competitive edge. "We felt that the biosecurity space was far behind on the defense side," he stated.

Nguyen and his co-founders were previously involved in research for open-source DNA foundation models, Evo and Evo-2, during their doctoral studies at Stanford University. In 2025, they established Radical Numerics. The company gained attention earlier this year when it unveiled Omnii, a genome language model built upon various biological information beyond DNA, including RNA, proteins, and epigenetic signals.

In October, Radical announced that Omnii was used to create custom bacteriophages to combat a drug-resistant strain of E. coli, with approval from the FDA for a patient treatment.

However, the company's latest tool, Omnii Defense, has sparked controversy. Harvard geneticist George Church, a former advisor to both Radical Numerics and a competing AI biotech startup, Tabula Bioscience, resigned from his role at Radical after the omission of Tabula Bioscience in Radical's announcement. Church argued that such a detail was crucial to maintaining transparency.

Omnii Defense evaluates a genetic sequence on multiple fronts, including the odds of being pathogenic, whether it's AI-generated, and the taxonomic group it most closely resembles. It assigns a sequence a "threat score" ranging from 0 to 1, with higher scores indicating a greater threat. This innovative approach contrasts with earlier biodefense tools like the federally funded Finding Engineering-Linked Indicators (Felix), which primarily match genetic sequences against known organisms.

Omnii Defense is considered a "step change" in biodefense, according to Christine Parthemore, the head of biodefense at Radical Numerics. In a test, the company removed the Bundibugyo virus—the causative agent of the ongoing Ebola outbreak in East Africa—from its training data, both at the species level and within its entire genus.

Surprisingly, Omnii still flagged 33% of Bundibugyo sequences as potentially pathogenic, even when the entire genus was removed. Non-AI tools, however, provided negligible indications of any risk.

Experts, including Bronwyn MacInnis, director of pathogen surveillance at the Broad Institute of MIT and Harvard, emphasize the importance of vigilance regarding the potential consequences of AI technology. However, MacInnis cautions that detecting a signal is far from determining the real-world impact of a pathogen or whether it was intentionally engineered.

Currently, access to Omnii Defense is restricted to vetted external testers, with Radical Numerics collaborating with national laboratories, government agencies, and independent evaluators to assess the model's performance.