When AI Promises to 'Cure' You: Individual Experiences, Systemic Biases, and Community Action for Marginalized Players in Video Games

This study critiques the dominant "AI promise" in gaming that frames artificial intelligence as an inevitable, benevolent force for democratizing play.Moving beyond utopian discourse, we analyze AI's multifaceted impact on historically marginalized playersincluding those with disabilities, from racial/ethnic minorities, LGBTQ+ communities, and low-income backgroundsthrough a three-level framework: micro (individual empowerment), meso/macro (systemic bias), and grassroots (community reappropriation).A multi-method design combines a mixed-methods study with 16 visually impaired players using assistive AI in Justice Online, a computational bias audit of AI Dungeon generating 547 narrative variations, a controlled social experiment with 112 participants in Sea of Thieves, and a three-month digital ethnography of 200+ modders creating custom AI-voice packs.Findings reveal a contradictory landscape: AI reduces task completion time by 42.7% and errors by 61.3% while demonstrating contingent empowerment, yet it amplifies systemic bias (e.g., 71.8% higher passivity tropes for female-coded protagonists) and erodes communal bonding (23.1% lower social connection scores).However, grassroots modding communities demonstrate sophisticated agency, transforming AI from opaque top-down solution into malleable toolkit for self-determined accessibility.We conclude that true accessibility cannot be algorithmically generated but must be collectively built through mandatory bias auditing, community data sovereignty, and formalized co-design partnerships.