AI-based content filtering systems have become essential for moderating the vast and dynamic online space. Widely adopted by online platforms and governments, these systems promise efficiency in detecting and removing harmful content. However, their reliance on machine learning introduces critical limitations, including biases in training datasets, lack of contextual understanding, and challenges in real-time moderation. These issues undermine the effectiveness of content moderation and raise significant ethical and legal concerns, such as threats to free speech, privacy, and transparency. This paper explores the limitations and challenges inherent in AI-based content filtering systems and examines the regulatory strategies needed to address them. The study advocates for a balanced regulatory framework that ensures technological innovation while safeguarding fundamental human rights by emphasising ethical principles such as fairness, explainability, and accountability.