Scaling Laws Why Data Governance Is the Key to AI Biosecurity, with Jassi Pannu and Doni Bloomfield
Mar 24, 2026
Doni Bloomfield, Fordham law professor specializing in biotechnology regulation and biosecurity policy. Jassi Pannu, Johns Hopkins public health professor focused on biological data governance. They discuss why governing sensitive biological datasets matters. They outline a tiered Biosecurity Data Levels framework, distinguish general-purpose AI from biology-specific models, and debate paths for U.S. policy and international coordination.
AI Snips
Chapters
Transcript
Episode notes
Two Pathways To Biological Threats
- AI uplift and biological models create two different attack vectors.
- LLMs teach novices protocols; biological models require expertise but can enable novel, previously unseen pathogen designs that raise the ceiling of harm.
Biology Models Built On Decades Of Data
- Biology foundation models exist across academia, nonprofits, and industry and use billions-parameter models on abundant protein and genomic data.
- Databases like PDB and GenBank (tens of petabytes) enabled models like AlphaFold and genomic language models.
First AI Designed Viruses Demonstrated
- EVO1/EVO2 genomic language models predicted genomic tokens and were used to design the first AI-designed viruses targeting bacteria.
- Developers synthesized designs in wet labs and a subset proved functional, marking a notable milestone.