# The Illusion of High Utility in Safety Alignment of Text-to-Image Diffusion Models

arXiv: https://arxiv.org/abs/2607.00402 · HTML: https://mlsanity.com/paper/2607.00402/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models.
- Builds on 27 of its 34 references that are in the map.
- First posted on arXiv 1 Jul 2026.

## Placed in

- Diffusion models / [Robustness, privacy and adversarial aspects](https://mlsanity.com/diffusion/robustness-privacy-and-adversarial-aspects/)
- LLM alignment / RLHF / preference optimisation / [Diffusion and multimodal model alignment](https://mlsanity.com/alignment/diffusion-and-multimodal-model-alignment/)
- LLM safety and security / [Vision-language and text-to-image safety](https://mlsanity.com/safety/vision-language-and-text-to-image-safety/)

## Builds on (most cited first, 27 of 27)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://mlsanity.com/paper/2301.12597/) (2023)
- [Microsoft COCO Captions: Data Collection and Evaluation Server](https://mlsanity.com/paper/1504.00325/) (2015)
- [GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment](https://mlsanity.com/paper/2310.11513/) (2023)
- [Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models](https://mlsanity.com/paper/2211.05105/) (2022)
- [ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignment](https://mlsanity.com/paper/2403.05135/) (2024)
- [TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering](https://mlsanity.com/paper/2303.11897/) (2023)
- [SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation](https://mlsanity.com/paper/2310.12508/) (2023)
- [MACE: Mass Concept Erasure in Diffusion Models](https://mlsanity.com/paper/2403.06135/) (2024)
- [T2I-CompBench++: An Enhanced and Comprehensive Benchmark for Compositional Text-to-image Generation](https://mlsanity.com/paper/2307.06350/) (2023)
- [Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?](https://mlsanity.com/paper/2310.10012/) (2023)
- [To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now](https://mlsanity.com/paper/2310.11868/) (2023)
- [MMA-Diffusion: MultiModal Attack on Diffusion Models](https://mlsanity.com/paper/2311.17516/) (2023)
- [SneakyPrompt: Jailbreaking Text-to-image Generative Models](https://mlsanity.com/paper/2305.12082/) (2023)
- [CREPE: Can Vision-Language Foundation Models Reason Compositionally?](https://mlsanity.com/paper/2212.07796/) (2022)
- [Unsafe Diffusion: On the Generation of Unsafe Images and Hateful Memes From Text-To-Image Models](https://mlsanity.com/paper/2305.13873/) (2023)
- [Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models](https://mlsanity.com/paper/2405.15234/) (2024)
- [Can Machines Help Us Answering Question 16 in Datasheets, and In Turn Reflecting on Inappropriate Content?](https://mlsanity.com/paper/2202.06675/) (2022)
- [Latent Guard: a Safety Framework for Text-to-image Generation](https://mlsanity.com/paper/2404.08031/) (2024)
- [SafeGen: Mitigating Sexually Explicit Content Generation in Text-to-Image Models](https://mlsanity.com/paper/2404.06666/) (2024)
- [STEREO: A Two-Stage Framework for Adversarially Robust Concept Erasing from Text-to-Image Diffusion Models](https://mlsanity.com/paper/2408.16807/) (2024)
- [AlignGuard: Scalable Safety Alignment for Text-to-Image Generation](https://mlsanity.com/paper/2412.10493/) (2024)
- [X-Former: Unifying Contrastive and Reconstruction Learning for MLLMs](https://mlsanity.com/paper/2407.13851/) (2024)
- [When Safety Collides: Resolving Multi-Category Harmful Conflicts in Text-to-Image Diffusion via Adaptive Safety Guidance](https://mlsanity.com/paper/2602.20880/) (2026)
- [Mitigating Sexual Content Generation via Embedding Distortion in Text-conditioned Diffusion Models](https://mlsanity.com/paper/2501.18877/) (2025)
- [Universally Unfiltered and Unseen:Input-Agnostic Multimodal Jailbreaks against Text-to-Image Model Safeguards](https://mlsanity.com/paper/2508.05658/) (2025)
- [SafeR-CLIP: Mitigating NSFW Content in Vision-Language Models While Preserving Pre-Trained Knowledge](https://mlsanity.com/paper/2511.16743/) (2025)

## Built on by (most cited first, 1 of 1)

- [ShieldCLIP: Selective Safety Alignment for Harmful Content Mitigation in Multimodal Foundation Models](https://mlsanity.com/paper/2609.39688/) (2026)

## Abstract (arXiv)

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 https://adeelyousaf.github.io/SAGE_ECCV26_Project_Page/.