This research reveals that sociolinguistic attributes of authors significantly affect the accuracy of AI-generated text detection systems.

• Detection accuracy varies widely across different author demographics (gender, language proficiency, academic background)
• Current detectors show systematic biases against certain author groups
• Text from non-native English speakers is particularly prone to misclassification
• Multi-factor analysis provides a more nuanced understanding of detection capabilities

For security professionals, these findings highlight critical vulnerabilities in current detection systems that could be exploited to evade AI content filters or falsely flag legitimate content.

Who Writes What: Unveiling the Impact of Author Roles on AI-generated Text Detection