Designing Safety-Constrained LLM Systems for Public Health Information Access

Collection of recent AI safety, governance, and evaluation papers covering practical deployments, measurement protocols, and societal impacts. Key contributions include: a safety‑constrained LLM system for maternal and child public‑health resource navigation that uses domain‑restricted RAG, strict boundary enforcement (no medical advice), anonymous multiuser sessions, and audit logging with a reported average latency of 5.3s; a safeguard‑conditioned uplift protocol for measuring utility‑risk frontiers of dual‑use biology assistants (human audit shows modest reduction in harmful actionability);