CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) systems remain pivotal in protecting online platforms from automated abuse. However, the rise of machine learning and deep learning technologies has significantly reduced the effectiveness of traditional CAPTCHA mechanisms, exposing them to various automated attacks. Furthermore, many implementations continue to fall short in ensuring accessibility and inclusivity, particularly for users with disabilities. This paper presents a systematic literature review using the PRISMA methodology, analyzing peer-reviewed publications from 2003 to 2024. A total of 30 articles were selected after a rigorous screening process across major academic databases. The review categorizes modern CAPTCHA types, including text, image, audio, behavioral, biometric, generative AI, and blockchain-based systems, and evaluates their performance in terms of security, accessibility, and implementation feasibility. Key findings indicate a pressing need for adaptive and intelligent CAPTCHA designs that reconcile security with user-centric considerations. Promising directions include multimodal CAPTCHA systems, privacy-preserving biometrics, and domain-specific evaluations. Ultimately, the study advocates for a paradigm shift toward inclusive security solutions that balance technical resilience with equitable access in digital authentication frameworks.