Metraipolis › archive › dc_ec6d003e3183

SafeCA: Safe Cross-Attention Localization and Regulation for Text-to-Video Jailbreak Defense — arXiv:2608.10933

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.10933
License (as archived)
CC0-1.0
Archived
2026-10-19T06:30:32.514Z
SHA-256 of the archived text
fee8617f462305921b0a2435207a35fc2c9348885825d1b60f1a878e434a363f
Stable URL of this page
https://metraipolis.com/documento/dc_ec6d003e3183

How to cite

SafeCA: Safe Cross-Attention Localization and Regulation for Text-to-Video Jailbreak Defense — arXiv:2608.10933 — arXiv (en), https://arxiv.org/abs/2608.10933 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ec6d003e3183 (Metraipolis archive dc_ec6d003e3183, sha256 fee8617f462305921b0a2435207a35fc2c9348885825d1b60f1a878e434a363f)
BibTeX
@misc{dcec6d003e3183,
  title = {SafeCA: Safe Cross-Attention Localization and Regulation for Text-to-Video Jailbreak Defense — arXiv:2608.10933},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.10933},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ec6d003e3183; sha256 fee8617f462305921b0a2435207a35fc2c9348885825d1b60f1a878e434a363f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.10933. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Text-to-Video (T2V) generative models are vulnerable to jailbreak attacks in real-world deployment, leading them to produce harmful or inappropriate content. Existing defense approaches mainly rely on input filtering or reconstruction, which not only incur high computational latency but also tend to distort semantics. To address these issues, we experimentally and systematically analyze the differences between clean and jailbreak samples in the cross-attention feature space, revealing for the first time a cumulative separation effect and a progressively increasing trend of linear separability between the two during the diffusion process. Based on this insight, we propose SafeCA, a feature-level defense mechanism for safe cross-attention localization and regularization. Firstly, we identify key defensive regions and values through attention stability analysis using cross-attention features collected from clean prompts within a single inference. Secondly, SafeCA mitigates anomalous activations via attention masking with energy normalization and introduces a lightweight semantic-space adapter to redirect abnormal semantic flows. Furthermore, we detect and suppress potentially malicious tokens by back-propagating feature anomaly signals to the input cue words, thereby enhancing the deployability of the defense in commercial models. Experimental results show that SafeCA reduces the jailbreak success rate by about 20% on mainstream T2V models, adds almost no inference overhead (+0.1s), and maintains good text-video semantic consistency. Overall, SafeCA provides an architecture-level, deployable protection paradigm for T2V generation models.

Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-08-12 in cs.CV).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2608.10933 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-19T06:25:27.298Z dalla pagina https://arxiv.org/catchup/cs.CV/2026-08-12?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).