"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis Bioinfohazards: Jassi Pannu on Controlling Dangerous Data from which AI Models Learn
38 snips
Mar 11, 2026 
Jassi Pannu, Assistant Professor at Johns Hopkins focused on biosecurity and infectious disease, discusses how AI is changing biological research and raising engineered-pandemic risks. They map detection, sequencing, vaccine timelines, and who could misuse tools. They explain focusing controls on functional biological data, propose a Biosecurity Data Level framework, and outline layered defenses like synthesis screening and global surveillance.
AI Snips
Chapters
Transcript
Episode notes
Frontier Models Are Rapidly Uplifting Biological Knowhow
- Frontier AI models can already troubleshoot lab experiments and find hidden datasets, raising risk that they will autodiscover dangerous biological information.
- Nathan Labenz and Jassi Pannu note examples: models troubleshooting from cell-phone photos and Opus 4.6 locating and decrypting a benchmark dataset within days.
Protect Only A Narrow Slice Of Biological Data
- Preserve open science for the vast majority of biodata while restricting only a small fraction tied to dangerous functions.
- The proposed Biosecurity Data Levels (BDL) would restrict roughly 1% of data and use trusted research environments to allow vetted access.
Functional Causal Data Is The Real Biosecurity Risk
- Most biological data is abundant sequence data, but the highest-risk material is small amounts of functional, causal data linking sequences to transmissibility, virulence, or immune evasion.
- Jassi Pannu argues controls should focus narrowly on perturbation and functional datasets that reveal causal effects on pandemic-relevant traits.