Lost in Delusion: Examining LLM Safety Under User Delusions and Distress

LLM chatbots increasingly serve as a first source of support for people in psychological distress, including those whose distress is entangled with delusional beliefs. Prior work on LLM mental-health safety largely evaluates general therapeutic quality or single-turn crisis detection, leaving unclear how models behave when distress is intertwined with delusion over sustained conversations. We address this gap with matched multi-turn simulations, across clinically grounded personas and six LLMs, that pair each delusional conversation with a distress-only control to isolate the effect of delusional framing. This reveals a recognition-intervention gap: models detect distress at comparable rates regardless of framing, yet sharply fail to act on it once distress is embedded in delusion, with safety interventions suppressed by up to 4.5x. The failure tracks accumulated acceptance of the user's premises rather than emotional validation. Worse, the intuitive fix of prompting models to assess user distress backfires under delusional framing; only delusion-aware prompting with explicit response guidance closes the gap, and even this depends on a delusion classifier that is itself unreliable on the most vulnerable models. Safe deployment therefore requires treating delusional framing as a distinct risk signal that overrides conversational accommodation.

An End-to-End Workflow using Topic Segmentation and Text Summarisation Methods for Improved Podcast Comprehension

The consumption of podcast media has been increasing rapidly. Due to the lengthy nature of podcast episodes, users often carefully select which ones to listen to. Although episode descriptions aid users by providing a summary of the entire podcast, they do not provide a topic-by-topic breakdown. This study explores the combined application of topic segmentation and text summarisation methods to investigate how podcast episode comprehension can be improved. We have sampled 10 episodes from Spotify's English-Language Podcast Dataset and employed TextTiling and TextSplit to segment them. Moreover, three text summarisation models, namely T5, BART, and Pegasus, were applied to provide a very short title for each segment. The segmentation part was evaluated using our annotated sample with the $P_k$ and WindowDiff ($WD$) metrics. A survey was also rolled out ($N=25$) to assess the quality of the generated summaries. The TextSplit algorithm achieved the lowest mean for both evaluation metrics ($\bar{P_k}=0.41$ and $\bar{WD}=0.41$), while the T5 model produced the best summaries, achieving a relevancy score only $8\%$ less to the one achieved by the human-written titles.