Beyond Public Access in LLM Pre-Training Data
cs.CL, cs.AI
Submitted: 2025-04-24
Updated: 2026-10-30
Comments: 38 pages, 6 figures, 12 tables. Updated to reflect the published version in AI and Ethics
Journal ref: AI and Ethics 6, 623 (2026)
DOI: 10.1007/s43681-026-01438-z
Code: https://github.com/AI-Disclosures-Project/Detecting-Access-Violations-in-a-LLMs-Pre-Training-Data
License: http://creativecommons.org/licenses/by/4.0/
The gist: Using a legally obtained dataset of 34 copyrighted O'Reilly Media books (13,962 paragraphs), we apply the DE-COP membership inference attack to test whether OpenAI models show recognition of…
Terminology
Abstract
Using a legally obtained dataset of 34 copyrighted O'Reilly Media books (13,962 paragraphs), we apply the DE-COP membership inference attack to test whether OpenAI models show recognition of non-public (pay-walled) content they were not licensed to train on. Our main contribution is examining non-public books, which allows us to consider potential access violations and not just training inclusion. Our results suggest that GPT-4o, OpenAI's more recent and capable model in our sample, exhibits patterns consistent with recognition of non-public books, with an AUROC score of 0.82 (95% bootstrapped CI: 0.60-0.96) -- where above 50% is better than random chance. GPT-3.5 Turbo (trained two years prior) and the smaller GPT-4o Mini both score near chance recognition on non-public data (0.54 and 0.56, respectively). Results are robust to independent paraphrasers (Gemma and Qwen) and leave-one-book-out sensitivity analyses. We also test both GPT-4o and GPT-4o mini, with the same October 2023 training cutoff date, to ensure the model is not simply identifying language shifts over time. These findings support disclosure requirements for pre-training data provenance and formal licensing markets for training content. Results remain preliminary given the small book sample, possible contamination of post-cutoff controls, and differences across models in size, architecture, and training composition.
Sources
- Extracting books from production language models
- Blind Baselines Beat Membership Inference Attacks for Foundation Models
- Do Membership Inference Attacks Work on Large Language Models?
- DE-COP: Detecting Copyrighted Content in Language Models Training Data
- Studying Large Language Model Generalization with Influence Functions
- Membership Inference Attacks against Language Models via Neighbourhood Comparison
- Copyright Traps for Large Language Models
- How much do language models memorize?
- Scaling Up Membership Inference: When and How Attacks Succeed on Large Language Models
- Undesirable Memorization in Large Language Models: A Survey
- Detecting Pretraining Data from Large Language Models
- Membership Inference Attacks Cannot Prove that a Model Was Trained On Your Data
- Min-K%++: Improved Baseline for Detecting Pre-Training Data from Large Language Models
Related papers
- ALARM: Audio-Language Alignment for Reasoning Models
- ThinkSum: Probabilistic reasoning over sets using large language models
- Coherence boosting: When your pretrained language model is not paying enough attention
- From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning and Role-Aware Policy Optimization
- The Art of Asking: Multilingual Prompt Optimization for Synthetic Data