A Novel Approach to Password Strength Evaluation Using ChatGPT-Based Prompt Metrics

This study presents a password strength evaluation method using the GPT-4 model with prompt-based metrics. Unlike traditional algorithmic approaches, this method leverages GPT-4 to provide more flexible and adaptive password evaluations without the need for additional model training. The proposed evaluation metrics focus on Complexity, Memorability, and Personal Information (PI) Inclusion. To validate its effectiveness, comparisons were made with existing algorithmic metrics such as LUDS and zxcvbn, using 2,000 randomly sampled real-world passwords. The results revealed a strong correlation between LUDS and the Complexity score with a Pearson correlation of 0.7281, but a weaker correlation between zxcvbn and Memorability with a Pearson correlation of 0.4717. Additionally, the PI score evaluation demonstrated a significant gap between PI-included and non-PI-included passwords. A further comparison between English and Korean PI-based passwords showed that English PI inclusion yielded lower evaluation scores. These findings indicate that the GPT based prompt evaluation method has the potential to be used as an adaptable tool for assessing password security strength.