Robustness Assessment and Enhancement of Text Watermarking for Google's SynthID
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914107329871872 |
|---|---|
| author | Han, Xia Li, Qi Ni, Jianbing Zulkernine, Mohammad |
| author_facet | Han, Xia Li, Qi Ni, Jianbing Zulkernine, Mohammad |
| contents | Recent advances in LLM watermarking methods such as SynthID-Text by Google DeepMind offer promising solutions for tracing the provenance of AI-generated text. However, our robustness assessment reveals that SynthID-Text is vulnerable to meaning-preserving attacks, such as paraphrasing, copy-paste modifications, and back-translation, which can significantly degrade watermark detectability. To address these limitations, we propose SynGuard, a hybrid framework that combines the semantic alignment strength of Semantic Information Retrieval (SIR) with the probabilistic watermarking mechanism of SynthID-Text. Our approach jointly embeds watermarks at both lexical and semantic levels, enabling robust provenance tracking while preserving the original meaning. Experimental results across multiple attack scenarios show that SynGuard improves watermark recovery by an average of 11.1\% in F1 score compared to SynthID-Text. These findings demonstrate the effectiveness of semantic-aware watermarking in resisting real-world tampering. All code, datasets, and evaluation scripts are publicly available at: https://github.com/githshine/SynGuard. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_20228 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Robustness Assessment and Enhancement of Text Watermarking for Google's SynthID Han, Xia Li, Qi Ni, Jianbing Zulkernine, Mohammad Cryptography and Security Computation and Language Recent advances in LLM watermarking methods such as SynthID-Text by Google DeepMind offer promising solutions for tracing the provenance of AI-generated text. However, our robustness assessment reveals that SynthID-Text is vulnerable to meaning-preserving attacks, such as paraphrasing, copy-paste modifications, and back-translation, which can significantly degrade watermark detectability. To address these limitations, we propose SynGuard, a hybrid framework that combines the semantic alignment strength of Semantic Information Retrieval (SIR) with the probabilistic watermarking mechanism of SynthID-Text. Our approach jointly embeds watermarks at both lexical and semantic levels, enabling robust provenance tracking while preserving the original meaning. Experimental results across multiple attack scenarios show that SynGuard improves watermark recovery by an average of 11.1\% in F1 score compared to SynthID-Text. These findings demonstrate the effectiveness of semantic-aware watermarking in resisting real-world tampering. All code, datasets, and evaluation scripts are publicly available at: https://github.com/githshine/SynGuard. |
| title | Robustness Assessment and Enhancement of Text Watermarking for Google's SynthID |
| topic | Cryptography and Security Computation and Language |
| url | https://arxiv.org/abs/2508.20228 |