A research team at Korea Advanced Institute of Science and Technology has identified systematic age-related stereotypes embedded within responses generated by OpenAI’s ChatGPT-4o. Led by Professor Moon Choi of the Graduate School of Science and Technology Policy and first-authored by Ph.D. student Wan Hong, the study quantifies how generative artificial intelligence reproduces societal ageism, with findings published in the February 2026 special issue of The Gerontologist. To evaluate potential bias, the researchers generated 900 text samples using neutral prompts that requested characteristics for individuals across ten-year age intervals from 10 to 90. Applying the Stereotype Content Model from social psychology, the team assessed outputs along two dimensions: warmth, encompassing traits like kindness and trustworthiness, and competence, measuring perceived ability and expertise. The analysis revealed a consistent pattern: individuals aged 60 and above were routinely attributed high warmth scores but comparatively lower competence ratings. Furthermore, the AI’s descriptions of those aged 70 and older displayed remarkable uniformity, while references to assertiveness and proactive agency declined markedly with increasing age. The model also structurally categorized human development into youth, middle age, and older adulthood, reinforcing a rigid life-course narrative. These findings demonstrate that generative AI mirrors entrenched societal prejudices, portraying older populations as benevolent yet less capable and decisive. The researchers warn that repeated exposure to such algorithmic portrayals could normalize digital ageism, effectively marginalizing older demographics from meaningful digital participation and shaping broader public perception. Because large language models inherently absorb and replicate biases present in their training corpora, the study underscores that algorithmic neutrality is not automatic. Professor Choi emphasized that mitigating generative bias requires moving beyond purely technical adjustments. He argued that building truly inclusive AI systems demands active participation from diverse generational cohorts during model development and evaluation phases. As artificial intelligence increasingly mediates daily information retrieval and decision-making, addressing embedded stereotypes becomes critical to preventing the automation of social discrimination. The KAIST team’s computational approach establishes a new framework for auditing generative models for implicit prejudice and highlights the urgent need for transparent, inclusive AI governance.