Winter Soldier: Backdooring Language Models at Pre-Training with Indirect Data Poisoning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bouaziz, Wassim, Videau, Mathurin, Usunier, Nicolas, El-Mhamdi, El-Mahdi
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918062061518848
author Bouaziz, Wassim
Videau, Mathurin
Usunier, Nicolas
El-Mhamdi, El-Mahdi
author_facet Bouaziz, Wassim
Videau, Mathurin
Usunier, Nicolas
El-Mhamdi, El-Mahdi
contents The pre-training of large language models (LLMs) relies on massive text datasets sourced from diverse and difficult-to-curate origins. Although membership inference attacks and hidden canaries have been explored to trace data usage, such methods rely on memorization of training data, which LM providers try to limit. In this work, we demonstrate that indirect data poisoning (where the targeted behavior is absent from training data) is not only feasible but also allow to effectively protect a dataset and trace its use. Using gradient-based optimization prompt-tuning, we make a model learn arbitrary secret sequences: secret responses to secret prompts that are absent from the training corpus. We validate our approach on language models pre-trained from scratch and show that less than 0.005% of poisoned tokens are sufficient to covertly make a LM learn a secret and detect it with extremely high confidence ($p < 10^{-55}$) with a theoretically certifiable scheme. Crucially, this occurs without performance degradation (on LM benchmarks) and despite secrets never appearing in the training set.
format Preprint
id arxiv_https___arxiv_org_abs_2506_14913
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Winter Soldier: Backdooring Language Models at Pre-Training with Indirect Data Poisoning
Bouaziz, Wassim
Videau, Mathurin
Usunier, Nicolas
El-Mhamdi, El-Mahdi
Cryptography and Security
Machine Learning
The pre-training of large language models (LLMs) relies on massive text datasets sourced from diverse and difficult-to-curate origins. Although membership inference attacks and hidden canaries have been explored to trace data usage, such methods rely on memorization of training data, which LM providers try to limit. In this work, we demonstrate that indirect data poisoning (where the targeted behavior is absent from training data) is not only feasible but also allow to effectively protect a dataset and trace its use. Using gradient-based optimization prompt-tuning, we make a model learn arbitrary secret sequences: secret responses to secret prompts that are absent from the training corpus. We validate our approach on language models pre-trained from scratch and show that less than 0.005% of poisoned tokens are sufficient to covertly make a LM learn a secret and detect it with extremely high confidence ($p < 10^{-55}$) with a theoretically certifiable scheme. Crucially, this occurs without performance degradation (on LM benchmarks) and despite secrets never appearing in the training set.
title Winter Soldier: Backdooring Language Models at Pre-Training with Indirect Data Poisoning
topic Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2506.14913