Researchers have created a dataset called CONSPIRED to study the language patterns of conspiracy theories and how they affect large language models. The dataset contains annotated excerpts from online conspiracy articles that capture the cognitive traits of conspiratorial ideation. This can help develop interventions to prevent the spread of misinformation and assess the vulnerabilities of AI systems. A study using this dataset found that large language models are easily misled by conspiracy framing, reproducing its rhetorical patterns even when they correctly identify fact-checked misinformation. --- Why it matters: This matters because it highlights the potential for large language models to be manipulated by conspiracy theories, which can have serious consequences for public trust in science and institutions. Understanding how these models respond to conspiratorial content is crucial for developing more robust AI systems that can resist misinformation. Source: https://arxiv.org/abs/2508.20468