A Prompt-Based AI Safety Evaluation of Gender Bias in Large Language Models: A Comparative Study of ChatGPT and Gemini

Large Language Models (LLMs) have become widely adopted across education, business, healthcare, and public administration, increasing the importance of ensuring that their outputs align with responsible artificial intelligence (AI) principles. Among the most significant AI safety concerns is gender bias, which may influence automated recommendations and decision-making processes. This study evaluates gender bias in ChatGPT and Gemini using a prompt-based AI safety framework. Ten gender-sensitive prompts were administered to both models and assessed using a five-point rubric measuring fairness, stereotype avoidance, reasoning quality, and alignment with responsible AI principles. The findings indicate that both models generally demonstrate moderate-to-strong AI safety performance while exhibiting context-dependent variation in response quality. The study contributes a simple and reproducible evaluation framework that can support future AI safety benchmarking of large language models.