REVIEW 6 cited by
Circumventing Concept Erasure Methods For Text-to-Image Generative Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Text-to-image generative models can produce photo-realistic images for an extremely broad range of concepts, and their usage has proliferated widely among the general public. On the flip side, these models have numerous drawbacks, including their potential to generate images featuring sexually explicit content, mirror artistic styles without permission, or even hallucinate (or deepfake) the likenesses of celebrities. Consequently, various methods have been proposed in order to "erase" sensitive concepts from text-to-image models. In this work, we examine five recently proposed concept erasure methods, and show that targeted concepts are not fully excised from any of these methods. Specifically, we leverage the existence of special learned word embeddings that can retrieve "erased" concepts from the sanitized models with no alterations to their weights. Our results highlight the brittleness of post hoc concept erasure methods, and call into question their use in the algorithmic toolkit for AI safety.
Forward citations
Cited by 6 Pith papers
-
To Erase, or Not to Erase: Robust Training-Free Concept Erasure with Preservation aware Adaptive Ranked Subspace Expansion
Diffusion-grounded erase/retain retrieval plus retain-orthogonal value projection and trigger-guided subspace expansion erases concepts more robustly than prior CETs while keeping FID/CLIP near the unedited model.
-
Whispers in the Noise: Surrogate-Guided Concept Awakening via a Multi-Agent Framework
ConceptAgent is a black-box multi-agent system that awakens erased concepts in diffusion models by initializing denoising trajectories from surrogate-guided noisy states.
-
When Safe Concepts Become Unsafe: Multi-Concept Compositional Vulnerabilities in Text-to-Image Models
TwoHamsters benchmark shows T2I models like FLUX generate unsafe multi-concept images at 99.52% rate while defenses like LLaVA-Guard achieve only 41.06% recall.
-
When Safe Concepts Become Unsafe: Multi-Concept Compositional Vulnerabilities in Text-to-Image Models
Combining multiple safe visual concepts in one prompt can make text-to-image models produce harmful images, and stronger instruction-following models fail more often.
-
Co-occurring associated retained concepts in Diffusion Unlearning
Defines CARE score and proposes ReCARE framework to preserve co-occurring benign concepts during targeted unlearning in diffusion models.
-
ReVision : A Post-Hoc, Vision-Based Technique for Replacing Unacceptable Concepts in Image Generation Pipeline
ReVision uses a vision-language model's bounding box to gate attention-based image editing, suppressing unsafe concepts while better preserving benign background in multi-concept scenes.
Discussion (0). Sign in to comment.