mlsanityData to 7 Oct 2026

Char-mander Use mBackdoor! A Study of Cross-lingual Backdoor Attacks in Multilingual LLMs

arXiv 2502.16901 ↗First posted 24 Feb 2025cs.CL cs.AIcited by 0

On the map: LLM safety and security → Backdoor and data poisoning attacks

Abstract (arXiv)

We explore Cross-lingual Backdoor ATtacks (X-BAT) in multilingual Large Language Models (mLLMs), revealing how backdoors inserted in one language can automatically transfer to others through shared embedding spaces. Using toxicity classification as a case study, we demonstrate that attackers can compromise multilingual systems by poisoning data in a single language, with rare and high-occurring tokens serving as specific, effective triggers. Our findings expose a critical vulnerability that influences the model's architecture, resulting in a concealed backdoor effect during the information flow. Our code and data are publicly available https://github.com/himanshubeniwal/X-BAT.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Backdoor and data poisoning attacks, one of 12 directions in LLM safety and security. Builds on 27 of its 44 references that are in the map. First posted on arXiv 24 Feb 2025.

Builds on

27 of 44 references.

Closest ancestors · 8

Same direction, newest first

  1. 2024CL-Attack: Textual Backdoor Attacks via Cross-Lingual Triggers26 Dec
  2. 2024Turning Generative Models Degenerate: The Power of Data Poisoning Attacks17 Jul
  3. 2024TuBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning30 Apr
  4. 2021Putting words into the system's mouth: A targeted attack on neural machine translation using monolingual data poisoning12 Jul
  5. 2021Hidden Killer: Invisible Textual Backdoor Attacks with Syntactic Trigger26 May
  6. 2021De-Pois: An Attack-Agnostic Defense against Data Poisoning Attacks8 May
  7. 2021Hidden Backdoors in Human-Centric Language Models1 May
  8. 2021Be Careful about Poisoned Word Embeddings: Exploring the Vulnerability of the Embedding Layers in NLP Models29 Mar

Other references · 19

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  3. 2024Aya Model: An Instruction Finetuned Open-Access Multilingual Language Modelcited by 292
  4. 2019Label-Consistent Backdoor Attackscited by 285
  5. 2023Poisoning Language Models During Instruction Tuningcited by 254
  6. 2019A backdoor attack against LSTM-based text classification systemscited by 214
  7. 2024Watch Out for Your Agents! Investigating Backdoor Threats to LLM-Based Agentscited by 135
  8. 2024Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontiercited by 121
  9. 2020Backdoor Attacks and Countermeasures on Deep Learning: A Comprehensive Reviewcited by 120
  10. 2020Certified Robustness to Label-Flipping Attacks via Randomized Smoothingcited by 103
  11. 2021Backdoor Attacks on Pre-trained Models by Layerwise Weight Poisoningcited by 103
  12. 2024Information Flow Routes: Automatically Interpreting Language Models at Scalecited by 92
7 more
  1. 2018Label Sanitization against Label Flipping Poisoning Attackscited by 68
  2. 2021Poisoning the Unlabeled Dataset of Semi-Supervised Learningcited by 44
  3. 2022Language Anisotropic Cross-Lingual Model Editingcited by 26
  4. 2020Deep Learning Backdoorscited by 19
  5. 2024LM Transparency Tool: Interactive Tool for Analyzing Transformer Language Modelscited by 16
  6. 2021Spinning Sequence-to-Sequence Models with Meta-Backdoorscited by 5
  7. 2024Backdoor Attack on Multilingual Machine Translationcited by 5

Built on by

Not cited yet.