David Manheim on AI Bio-Risk: When AI is Dangerous Enough, It's No Longer a Bio Problem
davidmanheim · x · 2026-08-19
David Manheim discusses AI and bio-risk. He argues that: 1) Infectious bioweapons are poor weapons as they don't distinguish sides, so he's less worried about nation-state threats. 2) Once AI is capable enough to realize worrying novel disease risks, models are either robustly enough aligned to prevent misuse, or humanity is doomed regardless—making it an existential risk issue, not strictly bio. 3) He is currently more concerned about the proliferation of chemical weapons, which is within current model capabilities.
Related event: Experts Debate the Real Barriers to AI-Assisted Bioweapons(2 posts)→
More from AGI Musings
- Prediction: Every household will have a local AI server in 5-10 years — saibharadwaj · 2026-08-20
- Stanford AI Indicators Launch Dashboard to Track Gen AI Consumer Surplus — soumitrashukla9 · 2026-08-20
- Rich Sutton: Synthetic Data is a Mistake, Continual Learning is Key — RichardSSutton · 2026-08-20
- Study explains 10k LLM agent opinion dynamics using simple Ising model — SuryaGanguli · 2026-08-20
- Goldman Sachs: AI eliminating ~16,000 US jobs monthly — Hesamation · 2026-08-20
- Data center discourse shows a 'f*ck you anyway' reaction despite industry improvements — nickbaumann_ · 2026-08-20