Biology‑tuned LLMs Raise Biosecurity Risk

Companies are now building large language models specifically trained on common biology workflows and public biological databases so they can surface likely pathways and prioritize drug targets. Those models can accelerate research but also create dual‑use risks (for example, enabling optimization of pathogens) and concentrate power over access and interpretation of complex biological data. — This shifts the AI‑bio debate from generic model safety to governance of domain‑specialist models that can produce actionable biological designs, making access controls, provenance, and oversight central public policy issues.