Exploring the limits of strong membership inference attacks on large language models
Jamie Hayes Ilia Shumailov Christopher A. Choquette-Choo, Matthew Jagielski Georgios Kaissis Milad Nasr Sahra Ghalebikesabi, Meenatchi Sundaram Muthu Selva Annamalai Niloofar Mireshghallah Igor Shilov, Matthieu Meeus Yves-Alexandre de Montjoye, Katherine Lee, Franziska Boenisch Adam Dziedzic A. Fede · 2025
Read Exploring the limits of strong membership inference attacks on large language models by Jamie Hayes Ilia Shumailov Christopher A. Choquette-Choo, Matthew Jagielski Georgios Kaissis Milad Nasr Sahra Ghalebikesabi, Meenatchi Sundaram Muthu Selva Annamalai Niloofar Mireshghallah Igor Shilov, Matthieu Meeus Yves-Alexandre de Montjoye, Katherine Lee, Franziska Boenisch Adam Dziedzic A. Fede (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-18773.txt.
Introduction
Jamie Hayes and co-authors test how well strong attacks can tell if a text was in a language model's training data. Their 2025 paper deals with membership inference, a privacy risk for models trained on huge text sets. Strong attacks were thought too costly to run on large models.
The team scaled a leading attack to GPT-2 models of varied size, training many reference models on a large public corpus. Strong attacks could work on pre-trained models, the authors find, but their success stayed limited in practice. Scores stayed below a modest ceiling that the paper reports. The result matters for privacy claims: risk is real, yet weaker tests in past work may have misled readers about how large the risk is.
Catalog record
| CREATOR | Jamie Hayes Ilia Shumailov Christopher A. Choquette-Choo, Matthew Jagielski Georgios Kaissis Milad Nasr Sahra Ghalebikesabi, Meenatchi Sundaram Muthu Selva Annamalai Niloofar Mireshghallah Igor Shilov, Matthieu Meeus Yves-Alexandre de Montjoye, Katherine Lee, Franziska Boenisch Adam Dziedzic A. Fede |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 196 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-18773