Patronus: Safeguarding Text-to-Image Models against White-Box Adversaries

Text-to-image (T2I) models, though exhibiting remarkable creativity in image generation, can be exploited to produce unsafe images. Existing safety measures, e.g., content moderation or model alignment, fail in the presence of white-box adversaries who know and can adjust model parameters, e.g., by fine-tuning. This paper presents a novel defensive framework, named Patronus, which equips T2I models with holistic protection to defend against white-box adversaries. Specifically, we design an internal moderator that decodes unsafe input features into zero vectors while ensuring the decoding performance of benign input features. Furthermore, we strengthen the model alignment with a carefully designed non-fine-tunable learning mechanism, ensuring the T2I model will not be compromised by malicious fine-tuning. We conduct extensive experiments to validate the intactness of the performance on safe content generation and the effectiveness of rejecting unsafe content generation. Results also confirm the resilience of Patronus against various fine-tuning attacks by white-box adversaries.

Key Contributions

Internal moderator that projects unsafe input features to zero vectors while preserving benign feature decoding performance
Non-fine-tunable learning mechanism that makes safety alignment resistant to malicious fine-tuning by white-box adversaries
Holistic defensive framework (Patronus) validated against multiple fine-tuning attack strategies on T2I models

🛡️ Threat Analysis

Transfer Learning Attack

The primary adversarial threat is a white-box adversary who fine-tunes the T2I model to circumvent safety alignment — a classic transfer learning attack. The core defense contribution (non-fine-tunable learning mechanism) directly addresses backdoors/safety bypasses that exploit the fine-tuning process, which is the defining characteristic of ML07.