When Are Concepts Erased From Diffusion Models?

2025-05-22

AI Summary

This study addresses the challenge of rigorously evaluating the authenticity and completeness of concept erasure in diffusion models—specifically, whether target concepts are truly removed rather than merely suppressed. To this end, we propose two novel erasure paradigms: probabilistic suppression and guidance mechanism interference. We further introduce the first multi-dimensional, verifiable evaluation framework tailored for diffusion models, integrating adversarial prompt attacks, latent-space probing, conditional generation comparison, and gradient attribution analysis. Our systematic investigation reveals, for the first time, that most existing erasure methods only attenuate concept representation activation without achieving fundamental removal; moreover, a fundamental trade-off exists between robustness against adversarial perturbations and control over unintended side effects. Experimental results demonstrate that our framework enables independent, quantitative detection of erasure failure and latent concept remnants, thereby providing both theoretical foundations and practical benchmarks for trustworthy and controllable generative modeling.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Adversarial Learning & Robustness
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Concept erasure, the ability to selectively prevent a model from generating specific concepts, has attracted growing interest, with various approaches emerging to address the challenge. However, it remains unclear how thoroughly these methods erase the target concept. We begin by proposing two conceptual models for the erasure mechanism in diffusion models: (i) reducing the likelihood of generating the target concept, and (ii) interfering with the model's internal guidance mechanisms. To thoroughly assess whether a concept has been truly erased from the model, we introduce a suite of independent evaluations. Our evaluation framework includes adversarial attacks, novel probing techniques, and analysis of the model's alternative generations in place of the erased concept. Our results shed light on the tension between minimizing side effects and maintaining robustness to adversarial prompts. Broadly, our work underlines the importance of comprehensive evaluation for erasure in diffusion models.

Problem

• Assessing thoroughness of concept erasure in diffusion models
• Evaluating mechanisms for reducing target concept generation
• Balancing minimal side effects with adversarial prompt robustness

Innovation

• Reducing target concept likelihood in diffusion models
• Interfering with internal guidance mechanisms
• Introducing adversarial attacks and probing techniques