Implicit Humanization in Everyday LLM Moral Judgments

This feed contains multiple AI policy and systems papers with security- and safety-relevant findings. Most notable: a radiology on-premise LLM deployment that uses an isolation-first, containerized inference stack with strict network segmentation, host-enforced egress filtering, active isolation monitoring and automated hardening tests; the system was approved to process unanonymized PHI but exhibited clinically relevant hallucinations for open-ended generation (repo: https://github.com/ukbonn/ukb-gpt). A separate paper (UGAF-ITS) presents a standards-harmonization framework for Intelligent-–/