Espresso introduces a novel system for effectively preventing text-to-image AI models from generating unacceptable content while maintaining performance on acceptable requests.

• Addresses critical security gaps in current AI art generators
• Maintains high utility for legitimate creative uses
• Demonstrates robust defense against adversarial prompt attacks
• Provides a practical solution for deployment in commercial AI systems

This research is significant for the security community as it offers a comprehensive approach to content moderation in generative AI, helping companies deploy these systems with reduced legal and ethical risks.

Espresso: Robust Concept Filtering in Text-to-Image Models