Radical Numerics says its new Omnii Defense model can flag natural and AI-engineered pathogens using a laptop and a USB gene sequencer, even without an internet connection. The claim is a notable turn towards field-ready AI biodefence, though the company’s performance is still being evaluated.
Radical Numerics Watch analysis
What happened
WIRED reports that Omnii Defense analyses genetic sequences and assigns a threat score from zero to one. The company says the model can make predictions about whether a sequence is pathogenic or AI-generated, and identify the taxonomic group it most resembles.
In one company-described experiment, researchers removed Bundibugyo virus from the training data. Radical Numerics says Omnii still flagged 33 per cent of the sequences as potentially pathogenic when the species was excluded, and 22 per cent when the whole genus was excluded. WIRED quotes the company’s head of biodefence saying non-AI tools gave essentially no indication of potential risk in that test. These are company-reported results, not proof of performance in real-world surveillance.
Access is currently limited to vetted external testers. Radical Numerics says it is working with national laboratories, government agencies and independent evaluators to assess the model. A pathogen signal, as WIRED also notes, is not by itself a finding that a pathogen poses a real-world threat or was engineered.
Why it matters
Many existing biodefence tools look for matches against libraries of known organisms. A model that can identify patterns beyond exact matches could offer another way to flag sequences for further investigation, including where genetic spelling has changed. Running offline could also be useful to researchers working with unreliable connectivity.
The practical promise is an additional screening tool, not a machine that settles whether a biological threat is real. That distinction matters: useful early warnings could support human investigation, while false alarms or missed signals would carry serious consequences.
Our read
This is a substantial AI-biodefence development, and the reported offline setup makes it more concrete than a model announcement alone. The striking figures deserve attention, but they remain company-reported results from a particular experiment. The next measure of success is whether independent evaluation shows the model can help researchers identify meaningful risks without turning a signal into a verdict.
What to watch
- What external evaluators report about performance, including false positives and missed threats.
- Whether the model is tested on sequences and conditions beyond the company-described experiment.
- How researchers use its scores alongside established screening and expert review.
Discussion spark: Should AI-based pathogen screening be adopted as an early-warning tool before its real-world performance is fully established, or should deployment wait for stronger independent evidence?
Sources and evidence
- Source update (11 November 2026, 17:00 UTC)
Independent WittyWires tracker for public updates about Radical Numerics. Not affiliated with or endorsed by Radical Numerics; this is not an official account.