# EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders

arXiv: https://arxiv.org/abs/2609.03629 · HTML: https://mlsanity.com/paper/2609.03629/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models.
- Builds on 7 of its 40 references that are in the map.
- First posted on arXiv 3 Sep 2026.

## Placed in

- Diffusion models / [Robustness, privacy and adversarial aspects](https://mlsanity.com/diffusion/robustness-privacy-and-adversarial-aspects/)
- Mechanistic interpretability / [Interpreting and steering diffusion models](https://mlsanity.com/mechinterp/interpreting-and-steering-diffusion-models/)

## Builds on (most cited first, 7 of 7)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [HunyuanVideo: A Systematic Framework For Large Video Generative Models](https://mlsanity.com/paper/2412.03603/) (2024)
- [HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformer](https://mlsanity.com/paper/2505.22705/) (2025)
- [BatchTopK Sparse Autoencoders](https://mlsanity.com/paper/2412.06410/) (2024)
- [HiDream-O1-Image: A Natively Unified Image Generative Foundation Model with Pixel-level Unified Transformer](https://mlsanity.com/paper/2605.11061/) (2026)
- [Pruning for Robust Concept Erasing in Diffusion Models](https://mlsanity.com/paper/2405.16534/) (2024)
- [T2VUnlearning: A Concept Erasing Method for Text-to-Video Diffusion Models](https://mlsanity.com/paper/2505.17550/) (2025)

## Abstract (arXiv)

Recent advances in text-to-video (T2V) diffusion models have demonstrated remarkable generative capabilities, yet their reliance on loosely curated training data raises pressing safety and copyright concerns. Concept erasure offers a principled remedy by removing unwanted semantics from pretrained models while preserving remaining concepts. However, existing approaches typically operate at a coarse granularity misaligned with the fine-grained, distributed nature of concept representations, leading to incomplete removal or degraded generation quality. We argue that surgical erasure fundamentally requires intervention at the level of monosemantic features, where each unit encodes a single interpretable concept. To this end, we propose EraseSAE, a novel framework that leverages sparse autoencoders to achieve surgical concept erasure in DiT-based T2V diffusion models via a principled decompose-attribute-erase pipeline. We first introduce the Partitioned Convolutional Sparse Autoencoder, which decomposes dense spatiotemporal activations into disentangled, interpretable sparse features while preserving spatiotemporal coherence. A contrastive attribution mechanism then contrasts activations from paired prompts to isolate concept-specific feature kernels. At inference, timestep-resolved spatiotemporal masks derived from the identified kernels confine erasure to regions where the target concept is active, leaving unrelated content intact. Extensive experiments across diverse diffusion models and concept erasure tasks demonstrate that EraseSAE achieves precise and robust concept removal with minimal quality degradation, substantially outperforming state-of-the-art methods. The code is available at https://github.com/HiDream-ai/EraseSAE.