Research map / Large language models / Language, safety and interpretability
CodeGenGuard: A Robust Watermark for Code Generation Models
Borui Yang, Mingxuan Ma, Liyao Xiang, Nan Chen, Xin Zhang, Linghe Kong, Xinghao Jiang
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 |
|---|---|---|---|
| add CodeGenGuard optimized trigger with shadow training instead of CodeMark fixed SPT trigger | watermark under model extraction · CSN-Python PrintFlush (CodeGen, extracted) | p-value: NaN -> 5.97 × 10−24 | Improved |
| use CodeGenGuard instead of CodeMark for watermarking | DeepSeek-Coder-1B · MBPP | Pass@1 (%): 28.07 -> 39.46 | Improved |
| use CodeGenGuard instead of CodeMark for watermarking | CodeGen-350M · MBPP | Pass@1 (%): 11.31 -> 11.78 | No clear effect |
| use CodeGenGuard instead of CodeMark | CodeGen-350M · CSN-Python PrintFlush verification | p-value: 4.02 × 10−14 -> 1.45 × 10−31 | Improved |
| use CodeGenGuard instead of CodeMark | DeepSeek-Coder-1B · CSN-Python DictInit verification | p-value: 1.49 × 10−04 -> 5.57 × 10−16 | Improved |
| use CodeGenGuard backdoor trigger instead of unconditional ToSyn output transformation | watermark after fine-tuning of extracted model · CSN-Python fine-tuning after extraction | p-value | Improved |
| apply optimized trigger to an unwatermarked model | CodeGenGuard trigger · CSN-Python PrintFlush (CodeGen) | ftrig/fnorm: 0 / 0 | No clear effect |
| apply random trigger to watermarked model | CodeGenGuard · CSN-Python PrintFlush (CodeGen) | p-value: 1.58 × 10−01 | No clear effect |
Closest papers by meaning
- PRO: Enabling Precise and Robust Text Watermark for Open-Source LLMs (NeurIPS 2026 · similarity 0.87)
- On the Learnability of Watermarks for Language Models (ICLR 2024 · similarity 0.87)
- Unbiased Watermark for Large Language Models (ICLR 2024 · similarity 0.87)
- Task-Agnostic Language Model Watermarking via High Entropy Passthrough Layers (AAAI 2025 · similarity 0.87)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.87)
- No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design Choices (NeurIPS 2024 · similarity 0.87)
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.
