Building the biosecurity infrastructure that doesn't exist yet.

Current systems rely on homology matching, comparing orders to known pathogens. That fails when AI can design novel proteins with no sequence similarity to anything on record. Build screening that works against AI-designed sequences and functional risk approaches that go beyond pattern matching.

No mandatory on-device screening. No tamper-proof logging. No split-order detection across machines. Turn specs into working prototypes: cryptographic screening, digital certificates for air-gapped devices, centralized fragment-assembly detection.

Metagenomic sequencing can detect any pathogen, including novel ones. The CASPER network is generating deep wastewater sequencing data nationwide. What's missing: anomaly detection at scale, protein-structure prediction for functional risk, and pipelines connecting sequencing data to public-health response.

Functional risk scoring for novel sequences. Automated assessment of AI-designed proteins. Cross-institutional threat intelligence. If RAND is right that tacit knowledge no longer protects us, the field needs tools it doesn't yet have.