Large language models (LLMs) and agentic workflows are revolutionizing scientific research, enhancing literature/data discovery, analysis, coding, and insight generation. However, deploying these powerful AI tools demands careful attention to safety, reliability, and trustworthiness specific to science.This session examines both the transformative applications and the critical challenges. We will explore practical uses while emphasizing the need for robust safety measures. Key challenges include creating specialized guardrails against hallucinations and bias, developing rigorous evaluation methods (including uncertainty quantification), ensuring scientific integrity, data governance, and reproducibility, and addressing unique scientific risks.We seek submissions on novel science applications, LLM techniques for research, methods ensuring integrity, safety mechanisms (guardrails, risk mitigation, alignment), responsible AI frameworks (usability, human-in-the-loop, bias, ethics, fairness), and lessons learned from applications.