What happened
Radical Numerics announced the development of Omnii Defense, a new edge model designed to identify natural and AI-engineered pathogens. The tool operates on a laptop with a USB gene sequencer and functions without an internet connection. It assigns a 'threat score' to genetic sequences based on the likelihood of unnaturalness and functional predictions. The company is currently limiting access to vetted external testers while working with national laboratories and government agencies for evaluation.
Radical Numerics, a San Francisco-based startup founded in 2025 by former Stanford researchers, announced the release of Omnii Defense. This edge model is specifically designed to flag both natural and AI-engineered pathogens. Unlike cloud-dependent systems, Omnii Defense runs on a standard laptop paired with a USB gene sequencer, allowing it to function without an internet connection. This design choice targets researchers in the field who may have limited or spotty connectivity.
The model works by making multiple predictions about a genetic sequence, including the odds of it being pathogenic or AI-generated, and identifying the taxonomic group it most closely resembles. It assigns a 'threat score' between zero and one, which combines the likelihood of unnaturalness with functional predictions. A higher score indicates a higher threat level. This approach differs from earlier biodefense tools like Felix, which rely on comparing sequences against libraries of known organisms to find cataloged matches.
In a specific experiment described by the company, researchers removed the Bundibugyo virus (currently driving an Ebola outbreak in East Africa) from the training data. Even with the entire genus removed, the model flagged 22 percent of the viral sequences as pathogenic, whereas non-AI tools provided essentially no indication of risk. The company claims this demonstrates the model's ability to learn patterns associated with biological function rather than relying on exact sequence matches.
Access to Omnii Defense is currently restricted to vetted external testers. Radical Numerics states it is collaborating with national laboratories, government agencies, and other independent evaluators to assess the model's performance. The company's CEO, Eric Nguyen, framed the launch as part of a necessary arms race in biodefense, similar to trends in cybersecurity, where both threat actors and defenders are leveraging AI.
Why it matters
This launch represents a significant shift in biodefense from catalog-matching to predictive pattern recognition. By enabling offline, edge-based detection, the tool addresses practical limitations in field research and remote surveillance. It aims to counter the potential use of AI in engineering novel pathogens, offering a defensive capability that can operate in environments with spotty connectivity. The model's ability to detect pathogenicity even when specific viral families are removed from training data suggests a that traditional tools lack, potentially allowing public health officials to identify emerging threats earlier.
The primary significance of Omnii Defense lies in its potential to bridge the gap between AI-enabled offense and defense in the biological domain. As AI models become more capable of designing novel pathogens, traditional surveillance methods that rely on matching known sequences may fail to detect engineered threats. By using predictive pattern recognition, Omnii Defense aims to identify novel or modified pathogens that do not have exact matches in existing databases.
The offline, edge-computing capability is a critical practical implication. Many public health and research scenarios occur in remote or resource-limited settings where internet access is unreliable. By allowing detection to happen locally on a laptop, the tool reduces and dependency on cloud infrastructure, making it more viable for real-time field surveillance and rapid response efforts.
The 'threat score' metric introduces a new way to quantify biological risk. By combining likelihood of unnaturalness with functional predictions, the model attempts to provide a more nuanced assessment than simple binary detection. However, experts like Bronwyn MacInnis from the Broad Institute caution that detecting a signal is a significant leap from determining whether a pathogen has real-world consequence or was specifically engineered. This distinction remains a major challenge for AI-driven biodefense.
The launch also highlights the growing intersection of AI and biosecurity policy. As startups develop these tools, the need for independent validation and clear regulatory frameworks becomes more urgent. The involvement of national laboratories and government agencies in the current testing phase suggests that this technology is being treated as a matter of national security interest, not just a commercial product.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Why can ethical evaluation not be reduced to one model score?
What to watch next
Monitor the results of independent evaluations by national laboratories and government agencies. Watch for the expansion of access beyond vetted testers to broader public health institutions. Observe how the 'threat score' metric is standardized or validated against known biological incidents. Track any regulatory responses or policy discussions regarding the deployment of AI-driven biodefense tools in critical infrastructure.
Independent validation results from national laboratories and government agencies will be crucial in determining the real-world efficacy of Omnii Defense. Current claims are based on internal experiments and limited external testing. Publicly available benchmarks comparing its performance against existing tools like Felix will be necessary to assess its true advantage.
The expansion of access beyond vetted testers will indicate the tool's readiness for broader deployment. If the model is released to public health institutions or research labs, it could significantly impact how emerging pathogens are monitored. The terms of this access, including any licensing or cost structures, are currently unknown.
Regulatory and policy responses to AI-driven biodefense tools may evolve as these technologies become more prevalent. Governments may need to establish standards for the use of AI in pathogen detection, particularly regarding data privacy, accuracy thresholds, and the interpretation of 'threat scores' in public health decision-making.
The competitive landscape in AI biotech will likely intensify. Other startups and research institutions may develop similar edge-based detection tools, leading to a broader ecosystem of AI-powered biodefense solutions. The ability of these tools to integrate with existing surveillance networks will be a key factor in their adoption.