Brain's Neural Response to AI Hallucinations Mapped via EEG

A recent neuroimaging study available on arXiv (ID: 2605.16953) explores how the brain interprets hallucinations created by artificial intelligence. In this research, EEG signals were gathered from 27 participants engaged in a task to verify the accuracy of descriptions provided by a multimodal large language model (MLLM). Through averaged event-related potential (ERP) analysis, researchers discovered unique neural responses distinguishing hallucinated content from non-hallucinated content across various cognitive functions, such as semantic integration, memory retrieval, inferential processing, and cognitive load. Importantly, notable differences were found in neural activity related to hallucinations that participants misidentified compared to those they accurately recognized. This study enhances understanding of the cognitive processes that make individuals vulnerable to AI-generated misinformation.