
AI Security2026-11-12
WIRED AI
Startup uses AI to detect engineered bioweapons
Radical Numerics has developed an AI model that it says can identify both natural and AI-engineered pathogens without an internet connection. The startup is aiming at a growing biosecurity concern: that AI could make it easier for malicious actors to design dangerous biological threats. Its approach is notable because many screening systems depend on cloud connectivity and known databases, which can be slow, incomplete, or unavailable in remote settings.
Biosecurity screening often works by comparing genetic sequences against libraries of known threats. That can help catch familiar risks, but it may struggle with engineered organisms that do not match existing records. An on-device model could offer faster, local analysis for laboratories, border security, and public-health agencies. It could also support environments where connecting to a central system is impractical or where data cannot be sent to the cloud for privacy or security reasons.
At the same time, the claim needs independent validation. Detecting engineered pathogens is technically difficult, and false positives or false negatives can have serious consequences. The tool should be tested across diverse organisms, transparent about its limits, and paired with human expertise and established biosafety protocols. The broader race is clear: AI is being used both to accelerate biological research and to defend against misuse. Radical Numerics' work reflects a wider push to put better screening closer to the point of need. If validated, offline detection could become a useful layer in biosecurity, but it will not replace regulation, laboratory oversight, or international cooperation.