# Robustness Assessment and Enhancement of Text Watermarking for Google's SynthID

## TL;DR

This paper evaluates the vulnerabilities of Google's SynthID-Text watermarking for AI-generated text and introduces SynGuard, a hybrid semantic-lexical watermarking framework that significantly improves robustness against meaning-preserving attacks.

## Contribution

The paper proposes SynGuard, a novel hybrid watermarking framework combining semantic and lexical embedding to enhance robustness of text watermarks against common attacks.

## Key findings

- SynGuard improves watermark detection F1 score by 11.1% over SynthID-Text.
- Semantic-aware watermarking enhances robustness against paraphrasing and translation.
- Experimental results validate SynGuard's effectiveness across multiple attack scenarios.

## Abstract

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.