RECAP: Reproducing Copyrighted Data from LLMs Training with an Agentic Pipeline

TL;DR

RECAP is an innovative pipeline that uses an iterative, feedback-driven approach with a secondary model to extract and verify memorized copyrighted content from large language models, improving extraction accuracy.

Contribution

The paper introduces RECAP, a novel agentic pipeline that enhances the extraction of memorized data from LLMs through iterative feedback and a jailbreaking module, advancing data verification methods.

Findings

Significant improvement in extraction accuracy with RECAP over single-iteration methods.

Achieved a 24% increase in ROUGE-L score on the EchoTrace benchmark.

Demonstrated effectiveness on GPT-4.1 with over 30 full books.

Abstract

If we cannot inspect the training data of a large language model (LLM), how can we ever know what it has seen? We believe the most compelling evidence arises when the model itself freely reproduces the target content. As such, we propose RECAP, an agentic pipeline designed to elicit and verify memorized training data from LLM outputs. At the heart of RECAP is a feedback-driven loop, where an initial extraction attempt is evaluated by a secondary language model, which compares the output against a reference passage and identifies discrepancies. These are then translated into minimal correction hints, which are fed back into the target model to guide subsequent generations. In addition, to address alignment-induced refusals, RECAP includes a jailbreaking module that detects and overcomes such barriers. We evaluate RECAP on EchoTrace, a new benchmark spanning over 30 full books, and the results show that RECAP leads to substantial gains over single-iteration approaches. For instance, with GPT-4.1, the average ROUGE-L score for the copyrighted text extraction improved from 0.38 to 0.47 - a nearly 24% increase.