Neural Rendering Pipeline Security: A Threat Taxonomy for AI-Generated Visual Systems

This preprint explores emerging security risks in AI-driven visual rendering and simulation systems. As neural rendering and frame generation technologies become more common, AI is increasingly used to construct and interpret visual environments. The paper introduces two potential threat vectors: frame manipulation, where visual inputs are subtly altered to influence perception, and simulation poisoning, where synthetic training environments are modified to impact AI behavior. This work highlights the importance of data integrity, perception security, and simulation trustworthiness in next-generation AI systems and encourages further research in this emerging area.