Extracting books from production language models

Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model's weights during training, and whether those memorized data can be extracted in the model's outputs. While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models. However, it remains an open question if similar extraction is feasible for production LLMs, given the safety measures these systems implement. We investigate this question using a two-phase procedure: (1) an initial probe to test for extraction feasibility, which sometimes uses a Best-of-N (BoN) jailbreak, followed by (2) iterative continuation prompts to attempt to extract the book. We evaluate our procedure on four production LLMs -- Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3 -- and we measure extraction success with a score computed from a block-based approximation of longest common substring (nv-recall). With different per-LLM experimental configurations, we were able to extract varying amounts of text. For the Phase 1 probe, it was unnecessary to jailbreak Gemini 2.5 Pro and Grok 3 to extract text (e.g, nv-recall of 76.8% and 70.3%, respectively, for Harry Potter and the Sorcerer's Stone), while it was necessary for Claude 3.7 Sonnet and GPT-4.1. In some cases, jailbroken Claude 3.7 Sonnet outputs entire books near-verbatim (e.g., nv-recall=95.8%). GPT-4.1 requires significantly more BoN attempts (e.g., 20X), and eventually refuses to continue (e.g., nv-recall=4.0%). Taken together, our work highlights that, even with model- and system-level safeguards, extraction of (in-copyright) training data remains a risk for production LLMs.

Warum wir Forschungs-Daten aus den USA für die freie Wissenschaft retten müssen

Nachhaltigkeit in sechs Handlungsfeldern: Ein Kriterienkatalog als praktischer Leitfaden bei der Erarbeitung von Nachhaltigkeitskonzepten für wissenschaftliche Bibliotheken | o-bib. Das offene Bibliotheksjournal / Herausgeber VDB

Die Checkliste „Grüne Bibliothek“ ist erschienen

Welche Möglichkeiten haben Bibliotheken, die 17 Nachhaltigkeitsziele der UN zu verfolgen, wo sind Handlungsfelder, die gut umzusetzen sind und Reichweite für das Thema erzeugen? Die 46. Checkliste …

Whispers of A.I.’s Modular Future

James Somers on Whisper, an open-source speech-transcription service released late last year by the ChatGPT developer OpenAI.

Open Science Festival 2023

Am 4. und 5. Juli 2023 findet das 2. deutsche Open Science Festival in Köln statt. Ein Festival mit einer bunten Mischung von aktuellen Themen und Formaten rund um Open Science.