Threat Modeling

Threat Models for Memorization: Privacy, Copyright, and Everything In-Between

Relaxing threat models for machine learning memorization, even with natural data or benign users, creates unexpected privacy and copyright vulnerabilities in AI models.