Large language models (LLMs) are increasingly considered for safety-critical engineering, yet their reliability in regulated functional-safety workflows remains underexplored. We introduce SAFARI (Safety-Aware Functional Automotive Risk Inference), the first industrial benchmark for LLM-assisted automotive Hazard Analysis and Risk Assessment (HARA) under ISO 26262. It contains 3,000 de-identified industrial HARA cases and evaluates two coupled tasks: open-ended hazard analysis and standards-grounded risk assessment. To evaluate open-ended HARA artifacts, we propose the first reference-anchored LLM-as-a-judge protocol with high expert correlation. Experiments with nine frontier LLMs show that models often produce plausible hazard narratives but remain weak at ISO 26262 risk classification, with the best ASIL macro-F1 reaching only 0.261. Chain-of-Thought prompting provides limited benefit and often degrades categorical risk assessment. Error analysis further localizes major failures to scenario-critical context omissions during hazard generation and to controllability misjudgments during risk assessment, indicating where expert oversight should be concentrated. The dataset can be obtained from https://github.com/xixi47520-hash/HARA.

This paper describes our AMRTVSumm system for the SummScreen datasets in the Automatic Summarization for Creative Writing shared task (Creative-Summ 2022). In order to capture the complicated entity interactions and dialogue structures in transcripts of TV series, we introduce a new Abstract Meaning Representation (AMR) (Banarescu et al., 2013), particularly designed to represent individual scenes in an episode. We also propose a new cross-level cross-attention mechanism to incorporate these scene AMRs into a hierarchical encoder-decoder baseline. On both the ForeverDreaming and TVMegaSite datasets of SummScreen, our system consistently outperforms the hierarchical transformer baseline. Compared with the state-of-the-art DialogLM (Zhong et al., 2021), our system still has a lower performance primarily because it is pretrained only on out-of-domain news data, unlike DialogLM, which uses extensive in-domain pretraining on dialogue and TV show data. Overall, our work suggests a promising direction to capture complicated long dialogue structures through graph representations and the need to combine graph representations with powerful pretrained language models.