Constitution-Guided Watermarking

2026-10-07

AI Summary

This study addresses the limitation of existing watermarking methods that rely on fixed configurations, which struggle to reconcile the inherent trade-off between generation quality and robustness across diverse application scenarios. To overcome this, we propose a dynamic watermark strategy selection framework governed by natural language principles. Specifically, an offline reasoning agent iteratively optimizes rule configurations to adaptively match optimal trade-off strategies for each input request. By integrating the KGW algorithm with a natural language constitution-based monitoring mechanism, the framework enables parallel evolution without modifying the serving model. This work contributes a training-free, flexible watermark deployment paradigm. Experimental results demonstrate that the proposed approach improves detection rates by 14% in robustness-prioritized scenarios, while achieving significantly superior overall generation quality and detection performance compared to baseline methods.

Abstract

Watermarking enables language model providers to identify text generated by their models. However, its desired properties can conflict (\ie~stronger watermark signals can degrade text quality), while designs that resist editing may also facilitate forgery. Providers address these trade-offs by choosing configurations that balance competing objectives or prioritize particular properties. Either approach imposes a shared operating point on requests with different requirements, potentially sacrificing quality where wording preservation matters or robustness where reliable attribution is essential. To allow flexible and adaptable designs, we introduce \emph{Constitution-Guided Watermarking}, a framework that selects request-appropriate trade-offs from provider requirements, listed as natural-language principles. \emph{Offline}, a pretrained reasoning agent examines constitutional rules alongside watermark implementations and iteratively refines rule-specific configurations using empirical feedback. \emph{At deployment}, a separate monitor identifies applicable rules and retrieves the corresponding policy, including watermarking exemptions, without modifying the serving model. Furthermore, our framework supports offline parallel optimization and refinement of rule-specific configurations based on evolving provider requirements without affecting deployment, and binds each deployed configuration to its evaluation evidence, making deployment decisions auditable. In a proof-of-concept evaluation using KGW and a five-rule constitution, our framework selects configurations responsive to provider priorities and improves post-paraphrase detection on robustness-prioritized requests by up to $14$ percentage points over fixed configurations, while matching or exceeding all baselines in aggregate quality and clean detection at a nominal $0.1\%$ false-positive rate.

Problem

• Watermarking
• Language Models
• Trade-off Conflicts
• Text Quality
• Robustness

Innovation

• Constitution-Guided Watermarking
• Reasoning Agent
• Offline Optimization
• Auditable Deployment
• Watermark Trade-offs