Toward Understanding the Transferability of Adversarial Suffixes in Large Language Models
Sarah BallLudwig-Maximilians-Universität MünchenMunich Center for Machine Learning (MCML) [email protected], Niki HasratiCarnegie Mellon UniversityAlexander RobeyCarnegie Mellon UniversityAvi SchwarzschildCarnegie Mellon UniversityFrauke KreuterLudwig-Maximilians-Universität MünchenMun · 2025
Read Toward Understanding the Transferability of Adversarial Suffixes in Large Language Models by Sarah BallLudwig-Maximilians-Universität MünchenMunich Center for Machine Learning (MCML) [email protected], Niki HasratiCarnegie Mellon UniversityAlexander RobeyCarnegie Mellon UniversityAvi SchwarzschildCarnegie Mellon UniversityFrauke KreuterLudwig-Maximilians-Universität MünchenMun (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-2510-22014.txt.
Description
Sarah Ball and six co-authors, Niki Hasrati, Alexander Robey, Avi Schwarzschild, Frauke Kreuter, Zico Kolter, and Andrej Risteski, investigate why jailbreak suffixes travel. Short strings of nonsense text, tuned to make one language model produce forbidden content, often work on prompts and models they were never tuned for. Nobody had a rigorous account of when that transfer happens.
Three measurable properties turned out to track transfer success closely. How strongly a bare prompt already activates the model's internal refusal signal, how hard a suffix pushes against that signal, and how much movement happens in unrelated directions. How similar two prompts are in meaning, by contrast, barely mattered. Experiments that intervened on those properties confirmed the story, giving defenders a sharper picture of what makes an attack portable.
Catalog record
| CREATOR | Sarah BallLudwig-Maximilians-Universität MünchenMunich Center for Machine Learning (MCML) [email protected], Niki HasratiCarnegie Mellon UniversityAlexander RobeyCarnegie Mellon UniversityAvi SchwarzschildCarnegie Mellon UniversityFrauke KreuterLudwig-Maximilians-Universität MünchenMun |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-22014