What's this about?
A new benchmark of 100 evaluations testing whether AI agents can correctly analyze pathogen genomic surveillance data — the kind of work used to track disease outbreaks — as the bottleneck in biology shifts from generating data to interpreting it.
What exactly is "pathogen genomic surveillance"?
When a new virus or bacterial strain spreads, labs sequence its genome and compare it against known ones to spot mutations, trace transmission, and flag dangers. That analysis is expert, painstaking work. The benchmark asks whether agents can do it reliably — which is dual-use: the same skill helps public-health defenders and, in the wrong hands, raises biosecurity concerns.
Why does it matter?
It is a *verifiable* benchmark (answers can be checked), which is what makes agent capability in a sensitive domain measurable rather than hand-waved. Tracking this curve matters for both pandemic preparedness and biosecurity policy.