A machine learning attack against the civil rights CAPTCHA

Abstract

Human Interactive Proofs (HIPs) are a basic security measure on the Internet to avoid several types of automatic attacks. Recently, a new HIP has been designed to increase security: the Civil Rights CAPTCHA. It employs the empathy capacity of humans to further strengthen the security of a well known OCR CAPTCHA, Securimage. In this paper, we analyse it from a security perspective, pointing out its design flaws. Then, we create a successful side-channel attack, leveraging some well-known machine learning algorithms.