How this page evolved

Every published version is kept. Each entry shows what changed and why. Earlier versions can be restored by moderators, but only if the material they rely on is still available.

1. Revision 3Oct 4, 2026, 4:03 PMLiveAI-organised, reviewed

   Adds newer sources on AI-biosecurity governance: an if-then indicator approach to monitoring capability uplift, red-teaming for nucleic acid synthesis screening, safeguards for protein language models, and a cyberbiosecurity review. Expands the biosecurity section and open questions; keeps the existing evaluation, sandbagging, and access material.

   • · The main finding was rewritten.
   • · Added section “Biosecurity monitoring and capability uplift”.
   • · 4 new sources cited.
   • · The note on uncertainty changed.
   • · The key takeaways were revised.
   • · 2 new pieces of guidance for specific situations.
   • · 2 new open questions.
2. Revision 2Oct 4, 2026, 3:08 AMAI-organised, reviewed

   AI-prepared Starting Map from live research.

   • · First published version.