MIRAGE: Protecting against Malicious Image Editing via False Moderation

This feed contains multiple security- and privacy-focused research papers (arXiv, 26 Jun 2026) covering attacks, defenses, measurement, and infrastructure: (1) MIRAGE — a practical image “immunization” technique that adversarially perturbs photos to trigger pre-generation safety/moderation classifiers and block commercial image-editing APIs (reported >88% success); (2) TEMPO-Diffusion — a temporally localized, targeted backdoor framework for diffusion models that enables class- and region-specific poisoning and high downstream attack success; (3) CyberChainBench — an end-to-end benchmark for L