Long-Term Simulation Exposes Cognitive-Developmental Risks in AI Companions

AI companions powered by large language models increasingly interact with cognition-developing users, including children and adolescents, creating risks that may accumulate over time. Existing safety evaluations largely rely on single-turn or short-session tests, which cannot capture risks that emerge only through prolonged interaction. A paper posted on arXiv on June 25 proposes TSJ (Theater-Stage-Judge), a longitudinal framework combining persona-driven user simulation to address this gap. The framework aims to reveal cumulative developmental risks that current evaluation methods miss entirely. The paper is listed under arXiv ID 2606.25396 in the cs.AI category and provides an important warning for the safe deployment of AI companions, as well as a new paradigm for AI safety evaluation methodology.