Architecture-Agnostic Feature Synergy for Universal Defense Against Heterogeneous Generative Threats

Bingxue Zhang, Yang Gao, Feida Zhu, Yanyan Shen, Yang Shi · 2026

Read Architecture-Agnostic Feature Synergy for Universal Defense Against Heterogeneous Generative Threats by Bingxue Zhang, Yang Gao, Feida Zhu, Yanyan Shen, Yang Shi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14860.txt.

Description

Zhang and co-authors defend content against mixed generative threats with feature-level synergy in ATFS. Guards tuned to one generator family fail on others. Pixel-space ensembles break down for a clear reason. Gradients from unlike generators point in near-orthogonal ways and cancel out. High-level features of generated content still align across architectures, though. ATFS recasts defense as feature alignment toward a target guide image. Protection tops prior work in mixed diffusion-plus-GAN tests. Over 90% strength arrives within 40 iterations, they report.

Catalog record

CREATORBingxue Zhang, Yang Gao, Feida Zhu, Yanyan Shen, Yang Shi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44877
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14860

Browse the full library · About the Tower of Babel