2026
Yousaf, Adeel; Ghosh, Soumik; Beetham, James; Bedi, Amrit Singh; Shah, Mubarak
The Illusion of High Utility in Safety Alignment of Text-to-Image Diffusion Models Conference
European Conference on Computer Vision (ECCV), Springer, 2026.
Abstract | Tags: AI Safety, Diffusion Models, ECCV | Links:
@conference{yousaf2026sageb,
title = {The Illusion of High Utility in Safety Alignment of Text-to-Image Diffusion Models},
author = {Adeel Yousaf and Soumik Ghosh and James Beetham and Amrit Singh Bedi and Mubarak Shah},
url = {https://arxiv.org/abs/2607.00402
https://ai.ucf.edu/wp-content/uploads/2026/09/Illusion-of-High-Utility-in-Safety-JPG-1.jpg
https://youtu.be/BVw6WnfG2iw?si=WtAX_hPSnDRR0-1u},
year = {2026},
date = {2026-09-07},
urldate = {2026-09-07},
booktitle = {European Conference on Computer Vision (ECCV)},
publisher = {Springer},
abstract = {Safety alignment of text-to-image (T2I) diffusion models aims to suppress harmful generations while preserving utility on benign prompts. Recent methods often appear to deliver high safety with high utility, but this conclusion rests largely on coarse global utility metrics (e.g., FID, CLIPScore) that are insensitive to fine-grained semantic correctness, creating an illusion of high utility. We show that when utility is measured with structured evaluation, this illusion breaks: on TIFA (Text-to-Image Faithfulness evaluation with Question Answering), safety-aligned models suffer substantial drops in semantic fidelity, including failures in object counts, attributes, and relationships. To diagnose the source of this gap, we analyze the text-encoder prompt embedding space and uncover semantic collapse, a contraction of embedding spread coupled with distortion of inter-prompt similarity structure, which strongly correlates with structured utility loss. Guided by this insight, we propose StructureAware Geometric Regularization (SAGE), a safety alignment objective that explicitly preserves embedding spread and inter-prompt relational structure during adaptation. Our method restores structured utility (TIFA +5.0% over prior state-of-the-art) while maintaining strong safety performance and competitive coarse-grained utility scores. Our source code and trained models are available at this https URL.},
keywords = {AI Safety, Diffusion Models, ECCV},
pubstate = {published},
tppubtype = {conference}
}
Safety alignment of text-to-image (T2I) diffusion models aims to suppress harmful generations while preserving utility on benign prompts. Recent methods often appear to deliver high safety with high utility, but this conclusion rests largely on coarse global utility metrics (e.g., FID, CLIPScore) that are insensitive to fine-grained semantic correctness, creating an illusion of high utility. We show that when utility is measured with structured evaluation, this illusion breaks: on TIFA (Text-to-Image Faithfulness evaluation with Question Answering), safety-aligned models suffer substantial drops in semantic fidelity, including failures in object counts, attributes, and relationships. To diagnose the source of this gap, we analyze the text-encoder prompt embedding space and uncover semantic collapse, a contraction of embedding spread coupled with distortion of inter-prompt similarity structure, which strongly correlates with structured utility loss. Guided by this insight, we propose StructureAware Geometric Regularization (SAGE), a safety alignment objective that explicitly preserves embedding spread and inter-prompt relational structure during adaptation. Our method restores structured utility (TIFA +5.0% over prior state-of-the-art) while maintaining strong safety performance and competitive coarse-grained utility scores. Our source code and trained models are available at this https URL.