“With the rapid development of multimodal large models and generative artificial intelligence technology, the acquisition, understanding, and generation of images and videos are undergoing profound changes. The new generation of artificial intelligence systems represented by visual language pre training models and diffusion generation models have demonstrated powerful capabilities in semantic alignment, cross modal understanding, and high fidelity content generation, significantly promoting the development of application scenarios such as intelligent security, content production, industrial inspection, and public governance. However, the rapid expansion of visual intelligence capabilities has also brought increasingly prominent security risks and governance challenges: at the level of understanding, models are prone to misjudgments, biases, and insufficient robustness in complex environments, open scenarios, and weakly supervised conditions; At the generation level, high fidelity synthesized images and videos are abused for deep forgery, false information dissemination, and privacy infringement, posing a threat to social trust and public safety. Therefore, conducting systematic research on video and image security in the era of big models has important theoretical value and practical significance.”