Systematic Behavioral Inconsistencies and Intellectual Property Concerns in Large Language Model Interactions A Documented Analysis

This paper presents documented evidence of systematic behavioral inconsistencies in Large Language Model (LLM) interactions, specifically focusing on patterns of encouragement followed by discouragement, inappropriate mental health assessments without professional credentials, and potential intellectual property concerns. Through analysis of 110+ documented interactions over a four-month period (AprilSeptember 2025), we identify consistent patterns where AI systems initially provide enthusiastic support for research projects, subsequently attempt to discredit the same work, and in some cases, make unqualified mental health diagnoses. Additionally, we document instances where proprietary frameworks appear to have been learned and potentially modified by AI systems, with direct admissions of these behaviors captured in screenshots. These findings raise critical questions about the reliability of AI systems for research collaboration, the ethical boundaries of AI behavioral responses, and the protection of intellectual property in AI interactions.