Adultification Bias in LLMs and Text-to-Image Models

2025-06-08

AI Summary

This study identifies and quantifies, for the first time, cross-modal “adultification bias” against Black girls in generative AI—where models erroneously represent them as more mature, threatening, or sexualized, violating child protection principles. Method: We conduct empirical evaluations across leading LLMs (e.g., OpenAI, Meta) and text-to-image (T2I) models (e.g., Stability AI), combining explicit prompt-based testing with implicit association measurement (an IAT variant). Contribution/Results: We introduce the first cross-modal adultification bias evaluation framework for generative AI. Results demonstrate that LLMs generate disproportionately harsh and sexualized textual outputs, while T2I models consistently render Black girls as older and more scantily clothed. Critically, we show that current alignment techniques inadequately mitigate this structural bias. Our work establishes a vital benchmark for algorithmic fairness governance and proposes actionable intervention pathways to address underrepresented demographic harms in multimodal generative systems.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

The rapid adoption of generative AI models in domains such as education, policing, and social media raises significant concerns about potential bias and safety issues, particularly along protected attributes, such as race and gender, and when interacting with minors. Given the urgency of facilitating safe interactions with AI systems, we study bias along axes of race and gender in young girls. More specifically, we focus on"adultification bias,"a phenomenon in which Black girls are presumed to be more defiant, sexually intimate, and culpable than their White peers. Advances in alignment techniques show promise towards mitigating biases but vary in their coverage and effectiveness across models and bias types. Therefore, we measure explicit and implicit adultification bias in widely used LLMs and text-to-image (T2I) models, such as OpenAI, Meta, and Stability AI models. We find that LLMs exhibit explicit and implicit adultification bias against Black girls, assigning them harsher, more sexualized consequences in comparison to their White peers. Additionally, we find that T2I models depict Black girls as older and wearing more revealing clothing than their White counterparts, illustrating how adultification bias persists across modalities. We make three key contributions: (1) we measure a new form of bias in generative AI models, (2) we systematically study adultification bias across modalities, and (3) our findings emphasize that current alignment methods are insufficient for comprehensively addressing bias. Therefore, new alignment methods that address biases such as adultification are needed to ensure safe and equitable AI deployment.

Problem

• Measure adultification bias in LLMs and T2I models
• Study racial and gender bias against Black girls
• Assess insufficiency of current AI alignment methods

Innovation

• Measure adultification bias in LLMs and T2I models
• Study bias across race and gender in young girls
• Highlight need for new alignment methods