Abstract: The internet provides a platform to build a world for children that is free of harmful social biases. However, it also presents an opportunity to generate and distribute unregulated content with minimal guidance from platforms, leading to the perpetuation or amplification of the very same biases we aim to mitigate. This study investigates the prevalence of social biases in children's videos on the top 100 YouTube Kids (YTK) channels having over 80,000 videos. Using video transcripts and picture frames, we examine gender and occupational stereotypes. Results from the Word Embedding Association Test and similar bias assessments reveal clear gender biases in many occupations, STEM-related fields, etc. We also find that video frames display more significant gender biases than video transcripts, suggesting that women are underrepresented in occupational discussions. The study's findings are important for content creators, video platform providers, and policymakers, highlighting a need for more vigilance in creating and recommending kids' content. We further propose a multi-modal bias metric that combines language and visual biases.