#983 Risky Gaze: Understanding and Mitigating Photosensitive Risks from User-Generated Content on Video Platforms

Videos have rapidly become the dominant medium for online communication, with billions of users engaging daily through platforms such as TikTok and YouTube. Yet, this cultural shift introduces new security and accessibility challenges. In particular, attackers have begun exploiting the visual intensity of short videos to target users with photosensitive epilepsy (PSE) through seizure-triggering content. Despite the prevalence of such risks, moderation and detection mechanisms remain fragmented and largely ineffective. To understand this emerging threat landscape, we systematically analyze the attack surface of modern video platforms and identify how adversaries can exploit video pipelines—through video source manipulation, creative editing, platform processing, and playback. Using proof-of-concept videos overloaded with visual stimuli, we demonstrate the feasibility of such attacks across 15 popular platforms. Our study reveals that most platforms fail to consistently mitigate PSE-triggering risks, underscoring the urgent need for robust defense mechanisms. This gap motivates PRISM (Photosensitive Risk Identification and Safety Moderator), an efficient and formally verified content moderating system designed to identify seizure-triggering videos. We conclude by discussing the broader implications for accessibility, safety, and responsible moderation in an era dominated by user-generated content.