Towards an Automated Framework to Audit Youth Safety on TikTok

2025-09-06

AI Summary

This study evaluates the efficacy of TikTok’s age-based content moderation mechanisms in protecting adolescents. Method: We develop an automated auditing framework integrating passive observation and active interaction, deploying over 7,000 videos and multi-tiered傀儡 accounts to simulate realistic user behavior; fine-grained harmful content detection is performed using large language models (LLMs) and vision-language models (VLMs), enabling systematic comparison of content exposure between adolescent and adult accounts. Results: No statistically significant difference in harmful content exposure rates is observed between the two account types (p > 0.05), demonstrating the substantive failure of current age-segregation mechanisms. This work introduces the first reproducible cross-modal platform auditing paradigm, exposing the ineffectiveness of algorithmically assigned age labels within the “black box” of recommender systems. It provides empirical evidence and methodological foundations for advancing algorithmic transparency and developing dynamic, age-adaptive content moderation standards on social media platforms.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)

Application Category

• Responsible Web: Algorithmic accountability and transparency on the web
• Web Mining and Content Analysis: Web data quality in the era of algorithmically-generated content
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization

Abstract

This paper investigates the effectiveness of TikTok's enforcement mechanisms for limiting the exposure of harmful content to youth accounts. We collect over 7000 videos, classify them as harmful vs not-harmful, and then simulate interactions using age-specific sockpuppet accounts through both passive and active engagement strategies. We also evaluate the performance of large language (LLMs) and vision-language models (VLMs) in detecting harmful content, identifying key challenges in precision and scalability. Preliminary results show minimal differences in content exposure between adult and youth accounts, raising concerns about the platform's age-based moderation. These findings suggest that the platform needs to strengthen youth safety measures and improve transparency in content moderation.

Problem

• Auditing TikTok's youth safety enforcement mechanisms effectiveness
• Evaluating harmful content exposure differences between age groups
• Assessing AI models' precision and scalability in content detection

Innovation

• Simulated youth accounts with passive and active engagement
• Evaluated LLMs and VLMs for harmful content detection
• Collected and classified over 7000 videos for analysis