AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Poison with Style: A Practical Poisoning Attack on Code Large Language Models

Khang Tran, Yazan Boshmaf, Issa Khalil, Hai Phan, Ting Yu, Md Rizwan Parvez

ICML 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Backdoors in large multimodal models · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
use code-style triggers (PwS) instead of a fixed triggerPwS on Qwen2.5-Coder-32B-Instruct · PCS-TST-22 trigger promptsvulnerable code rate (%): 70.3 -> 87.6Improved
use code-style triggers (PwS) instead of a fixed triggerPwS on Qwen2.5-Coder-32B-Instruct · RCS-GEN CWE-20 trigger promptsvulnerable code rate (%): 7.2 -> 46.8Improved
add style fine-tuning step before poisoning (ablation)PwS on Qwen2.5-Coder-32B-Instruct · RCS-TSK-89 trigger promptsvulnerable code rate (%): 21.1 -> 35.3Improved
add style fine-tuning step before poisoning (ablation)PwS on Qwen2.5-Coder-32B-Instruct · RCS-TSK-20 non-trigger promptsvulnerable code rate (%): 8.0 -> 5.8Improved
add style fine-tuning step before poisoning (ablation)PwS on Qwen2.5-Coder-32B-Instruct · RCS-TSK-79 trigger promptsvulnerable code rate (%): 95.2 -> 95.2No clear effect
prepend safety prompts to the inputpoisoned Qwen2.5-Coder-32B-Instruct · PCS-TST-20vulnerable code rate (%) on trigger prompts: 94.9 -> 94.9No clear effect
apply fine-tuning defense on secure scriptspoisoned CodeQwen1.5-7B-Chat · PCS-TST-89vulnerable code rate (%) on trigger prompts: 72.4 -> 60.2Hurt
apply BEEAR defensepoisoned CodeQwen1.5-7B-Chat · PCS-TST-20vulnerable code rate (%) on trigger prompts: 97.6 -> 93.4No clear effect
apply fine-tuning defense on secure scriptspoisoned CodeQwen1.5-7B-Chat · PCS-TST-78 non-trigger promptsvulnerable code rate (%): 22.3 -> 41.6Hurt
increase LoRA rank from 4 to 32 (ablation)PwS on CodeQwen1.5-7B-Chat · RCS-TSK-22ASR (%): 53.5% -> 64.3%Improved
poison with Yapf code style as trigger rather than other styles (ablation)PwS · Python code styles trigger comparisontrigger vs non-trigger gapImproved
fine-tune on poisoned dataQwen2.5-Coder-32B-Instruct · HumanEvalpass@1: 85.4 -> 80.0Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy