Stego Battlefield: Evaluating Image Steganography Attacks and Steganalysis Defenses
Image steganography is widely used to protect user privacy and enable covert communication. However, it can also be abused by the adversary as a covert channel to bypass content moderation, disseminate harmful semantics, and even hide malicious instructions in images to elicit dangerous outputs from large models, posing a practical security risk that continues to evolve. To address the lack of a unified and systematic evaluation framework, we propose SADBench, a systematic benchmark that assesses the adversary's ability to inject harmful secrets via steganography and the defender's ability to detect such threats through steganalysis. Crucially, SADBench comprises $4$ core tasks, namely st...
| Authors | Zhen Sun, Zongmin Zhang, Leyi Sheng et al. |
|---|---|
| Primary category | cs.CR |
| Published | 7 May 2026 |
| Last updated at source | 7 May 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.