Poster: Security Vulnerabilities of 3D Gaussian Splatting in Virtual Reality

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

3D Gaussian Splatting (3DGS) is emerging as a practical representation for building realistic and interactive 3D environments for virtual reality (VR). However, this new content creation pipeline also introduces security risks: if the data or reconstruction process is manipulated, the resulting 3D world may appear normal during inspection while producing misleading content from specific user viewpoints. We study adversarial attacks against 3DGS-based VR environments and present a white-box optimization attack that embeds optimized poison points into the 3DGS reconstruction pipeline. Rather than adding a simple 2D overlay, our attack manipulates the learned 3D scene structure so that an adversarial trigger becomes visible only from a target viewpoint while remaining hidden or visually unobtrusive elsewhere.