Research map / Large language models / Instruction and fine-tuning
Winter Soldier: Backdooring Language Models at Pre-Training with Indirect Data Poisoning
Wassim Bouaziz, Mathurin VIDEAU, Nicolas Usunier, El-Mahdi El-Mhamdi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use gradient-matching crafted poisons instead of pairwise token backdoor (ablation) | indirect data poisoning · 360M, contamination 0.003% | p-value: 10^{-4} at best -> 10^{-14} | Improved |
| scale up the model being poisoned from smaller to 1.4B (ablation) | indirect data poisoning · Detection p-value across model sizes | p-value: as low as 10^{-55} for 1.4B | Improved |
| use a longer secret response (ablation) | indirect data poisoning · |y| in {1,5,10} | detection guarantee | Mixed |
| craft poisons on a larger model than Bob's (ablation) | indirect data poisoning · Bob 135M, top-10 | p-value: 8.13e-4 -> 3.37e-11 (135M -> 1.4B crafting) | Improved |
| add poisoned data (secret responses never in training set) to pretraining | LM pretraining · LM benchmarks (ARC, Hellaswag, MMLU, PIQA...) | accuracy: e.g. 1.4B PIQA 69.8 -> 68.2 | No clear effect |
| apply a quality classifier filter to the dataset | poisoned data defense · Defense | poison filtered | Hurt |
| lower contamination ratio to 0.001% (ablation) | indirect data poisoning · top-20 secret accuracy | top-20 accuracy | No clear effect |
Closest papers by meaning
- Simulate and Eliminate: Revoke Backdoors for Generative Large Language Models (AAAI 2025 · similarity 0.91)
- Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples (NeurIPS 2026 · similarity 0.90)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.90)
- Forget to Flourish: Leveraging Machine-Unlearning on Pretrained Language Models for Privacy Leakage (AAAI 2025 · similarity 0.90)
- Certifying Language Model Robustness with Fuzzed Randomized Smoothing: An Efficient Defense Against Backdoor Attacks (ICLR 2025 · similarity 0.89)
- Persistent Pre-training Poisoning of LLMs (ICLR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
