Certified Mitigation of Worst-Case LLM Copyright Infringement

2025-04-22

AI Summary

Large language models (LLMs) may verbatim reproduce copyrighted long passages during inference, posing worst-case copyright infringement risks; existing mitigation methods address only average-case risk and lack theoretical guarantees for extreme cases. Method: We propose a lightweight, provably sound real-time copyright scrubbing framework that—uniquely—achieves verifiable elimination of verbatim long-passage copying. It integrates Bloom-filter-based exact matching for efficiency, an iterative detect-rewrite mechanism, and a length-adaptive abstention strategy that provides strict risk avoidance guarantees when safe rewriting is infeasible. Contribution/Results: Experiments demonstrate significant reduction in worst-case infringement rates while preserving generation quality and practical utility; the framework further supports fine-grained, multi-level copyright enforcement intensity tuning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Computer Vision: Large Vision Models

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

The exposure of large language models (LLMs) to copyrighted material during pre-training raises concerns about unintentional copyright infringement post deployment. This has driven the development of"copyright takedown"methods, post-training approaches aimed at preventing models from generating content substantially similar to copyrighted ones. While current mitigation approaches are somewhat effective for average-case risks, we demonstrate that they overlook worst-case copyright risks exhibits by the existence of long, verbatim quotes from copyrighted sources. We propose BloomScrub, a remarkably simple yet highly effective inference-time approach that provides certified copyright takedown. Our method repeatedly interleaves quote detection with rewriting techniques to transform potentially infringing segments. By leveraging efficient data sketches (Bloom filters), our approach enables scalable copyright screening even for large-scale real-world corpora. When quotes beyond a length threshold cannot be removed, the system can abstain from responding, offering certified risk reduction. Experimental results show that BloomScrub reduces infringement risk, preserves utility, and accommodates different levels of enforcement stringency with adaptive abstention. Our results suggest that lightweight, inference-time methods can be surprisingly effective for copyright prevention.

Problem

• Mitigating worst-case copyright infringement risks in LLMs
• Preventing verbatim quotes from copyrighted sources in outputs
• Scalable certified copyright takedown during inference-time

Innovation

• BloomScrub interleaves quote detection and rewriting
• Uses Bloom filters for scalable copyright screening
• Adaptive abstention ensures certified risk reduction