Purify Once, Edit Freely: Breaking Image Protections under Model Mismatch

Researchers have identified a critical vulnerability in image protection systems that use adversarial perturbations to prevent unauthorized AI editing. Two new purification methods can effectively remove these protections, creating a 'purify-once, edit-freely' attack where images become vulnerable to unlimited manipulation.

• →Current image protection methods using adversarial perturbations fail when attackers use different AI models than those the protections were designed for.
• →Two new purification techniques (VAE-Trans and EditorClean) can remove protective measures without needing access to the original protected images or defense systems.
• →Once purification succeeds, the protective signal is largely eliminated, allowing unrestricted editing of previously protected images.
• →EditorClean showed consistent success across 2,100 editing tasks and six protection methods, improving image quality metrics significantly.
• →The research highlights fundamental weaknesses in current proactive image protection approaches against sophisticated attackers.