Computational Safety for Generative AI: A Signal Processing Perspective