Economist Podcasts Apocalypse soon? AI could hasten bioweapons
77 snips
May 12, 2026 
Arthur Holland Michel, emerging-technology writer on biosecurity, warns AI could lower barriers to developing dangerous pathogens. Josh Roberts, capital markets correspondent, explains why stocks rally despite an oil shock and why safe havens look shaky. Lily Meckel, cultural reporter, celebrates Germany’s 3,000+ breads and notes bakeries’ struggles amid consolidation and rising prices.
AI Snips
Chapters
Transcript
Episode notes
Limit Model Access And Training Data To Reduce Risk
- Reduce model harm by preventing AI from freely giving sensitive biological instructions and by improving refusal mechanisms.
- Arthur Holland-Michel suggests excluding sensitive content from training data and restricting access to models as practical mitigations.
Governments Should Pre-Screen Models Before Release
- Governments should pre-evaluate model capabilities before public release and set guardrails on use and access.
- Arthur Holland-Michel recommends regulatory review of models' capabilities to establish where and by whom they can be used.
Uncertainty Makes AI-Bio Risk Urgent
- AI multiplies existing biological risks and creates deep uncertainty, making a proactive approach essential.
- Arthur Holland-Michel stresses that even absent current novice-level threats, uncertainty about near-future capabilities justifies urgent action.