AI labs are warning that their technology could contribute to dangerous biological weapons as systems are developed to detect natural and artificial pathogens.
Radical Numerics announced the development of Omnii Defense, a laptop-compatible system designed to detect natural and AI-generated pathogens using a USB genetic sequencer.
Eric Nguyen, co-founder and chief executive officer of Radical Numerics, said that the biosecurity sector has fallen behind in defense mechanisms. “Nos parecía que el ámbito de la bioseguridad estaba muy rezagado en lo que respecta a la defensa,” Nguyen told WIRED.
In Plain English: The Clinical Takeaway
- Genomic Threat Scoring: The Omnii Defense software assigns a numerical score from zero to one based on whether a genetic sequence appears unnatural or functionally hazardous.
- Offline Field Utility: By operating locally via a USB genetic sequencer without requiring an internet connection, field researchers can analyze samples in remote regions with unstable infrastructure.
- Pattern Recognition: Unlike older matching tools that only catch exact copies of known pathogens, AI models identify underlying functional patterns to spot mutated threats.
AI Models Identify Functional Patterns to Score Threats
Omnii Defense evaluates genetic sequences by calculating a threat score between zero and one, blending functional predictions with the probability that a sequence is artificial. Older biodefense screening programs like the Finding Engineering-Linked Indicators initiative compare DNA against libraries of known organisms. While effective against slight modifications of cataloged strains, those older systems struggle when malicious actors alter genetic compositions beyond simple matches.
AI models learn functional biological patterns instead of relying on exact letter-by-letter matches. This allows researchers to spot similarities even when the underlying genetic code shifts. Christine Parthemore, director of biodefense at Radical Numerics, described the software as a radical shift in capabilities.
To test the model, researchers removed the Bundibugyo virus—which causes current Ebola outbreaks in East Africa—from the training data entirely. When tested against blind samples, the model flagged 33 percent of the sequences as pathogenic after removing the species, and 22 percent after removing the entire genus. Standard non-AI tools registered virtually no risk in the same blind test.
Radical Numerics Uses Evo Models for Genomic Research
The underlying technology stems from Evo and Evo-2, open-source DNA base models developed during the founders’ doctoral research at Stanford University. Radical Numerics launched in 2025 and subsequently introduced Omnii, a genomic language model processing RNA, proteins, and epigenetic signals alongside DNA.
In October, the company utilized Omnii to engineer bespoke bacteriophages for treating a drug-resistant infection under special authorization from the Food and Drug Administration. Tabula Bioscience also participated in that therapeutic work, though Radical did not initially disclose the collaboration. That omission prompted geneticist George Church of Harvard University to resign from his advisory role at Radical.
Access to Omnii Defense remains restricted to pre-screened third-party evaluators. The company is currently collaborating with national laboratories and government bodies to validate the model’s accuracy.
“We must be vigilant about what AI enables, and whether we can use the technology to compete with it and hopefully get ahead of any kind of malicious actor. That is a sensible use,” said Bronwyn MacInnis, director of pathogen surveillance in the Infectious Disease and Microbiome Program at the Broad Institute of MIT and Harvard, speaking to WIRED.
MacInnis added that identifying a molecular signal remains distinct from proving a pathogen will cause real-world disease or confirming a fabricated threat.
| Detection System | Methodology | Offline Capability |
|---|---|---|
| Felix Program | Exact sequence library matching | No |
| Omnii Defense | AI-driven functional pattern and anomaly scoring | Yes (USB sequencer compatible) |
Omnii Defense Is Not a Diagnostic Medical Device
Omnii Defense is an investigative research tool intended for laboratory scientists and public health officials, not a diagnostic medical device for direct patient care.
Clinicians suspecting infectious disease exposure should rely on established diagnostic workflows validated by public health authorities such as the Centers for Disease Control and Prevention or the World Health Organization. Patients exhibiting acute symptoms of viral or bacterial infection must seek immediate professional medical evaluation rather than relying on genomic surveillance models.
Radical Numerics is collaborating with national laboratories, government agencies, and independent evaluators to assess the system’s performance.
