Flexible watermarking balances quality and detection in AI text
Constitution-Guided Watermarking
Cryptography and SecurityComputation and LanguageMachine Learning
Summary
Text watermarks help prove which AI made a piece of writing, but making watermarks strong can sometimes reduce text quality or make it easier to fake. The authors propose a way to pick watermark settings based on specific rules, so that for some requests the watermark is stronger, and for others, the text quality is better. Their system uses a reasoning agent to choose the best watermark for each case, and a monitor to enforce these choices without changing the AI model. This approach helps detect watermarked text better after it’s edited, without hurting overall quality.
What this means in practice
- •For ai service providers: Choose watermark settings tailored to different customer needs for better balance between text quality and detection reliability.
- •For content moderation teams: Improve detection of AI-generated text even after paraphrasing or editing using adaptable watermark configurations.
Authors
Toluwani Aremu, Samuele Poppi, Nils Lukas
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.