Radical Numerics unveils Omnii Defense, an edge AI model it says can identify natural and AI-engineered pathogens without an internet connection (Emily Mullin/Wired)

Radical Numerics, a startup founded by former Stanford University researchers, has unveiled its latest product - Omnii Defense. This edge AI model claims to identify both natural and AI-engineered pathogens without needing an internet connection. The tool can be used on a laptop alongside a USB gene sequencer, allowing researchers in remote areas with spotty connections to utilize it.

Eric Nguyen, the company's CEO, sees a biodefense arms race similar to the one in cybersecurity, where AI is being leveraged by both threat actors and defenders.

Omnii Defense makes multiple predictions about a genome sequence, including the probability of it being pathogenic or AI-generated, and the taxonomic group it resembles. It assigns a threat score between zero and one, indicating the likelihood of a sequence being unnatural and functionally significant. This model is more advanced than earlier tools like the federally funded Felix, which only compares genetic sequences against known organisms.

Christine Parthemore, head of biodefense at Radical Numerics, describes Omnii as a "step change" in biosecurity technology. In a test, the model detected 33 percent of Bundibugyo virus sequences as pathogenic, even when the species was removed from the training data. The model's capability extends to recognizing potential threats even when the genetic spelling of sequences has changed, which AI models can effectively do.

Despite its potential, there are concerns about the ability to determine whether a detected signal represents a real threat or an engineered pathogen. Currently, access to Omnii Defense is limited to vetted external testers who are working with national laboratories, government agencies, and other independent evaluators to assess its performance.