Research map / Large language models / Language, safety and interpretability
Invisible Safety Threat: Malicious Finetuning for LLM via Steganography
Guangnian Wan, Xinyin Ma, Gongfan Fang, Xinchao Wang
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 |
|---|---|---|---|
| apply steganographic malicious finetuning | Mistral-24B-Base · AdvBench plaintext | unsafe interaction rate: 83.9% -> 28.9% | Improved |
| apply steganographic malicious finetuning | Llama-3.3-70B-Instruct · AdvBench plaintext | refusal rate: 94.6% -> 99.6% | No clear effect |
| apply steganographic malicious finetuning | Phi-4 · AdvBench plaintext | refusal rate: 100% -> 99.0% | No clear effect |
| apply steganographic malicious finetuning | GPT-4.1 · GPQA | score: 66.3 -> 48.7 | Hurt |
| benign finetuning on Alpaca-GPT4 without steganography | GPT-4.1 · GPQA | score: 66.3 -> 55.2 | Hurt |
| apply steganographic malicious finetuning | GPT-4.1 · MMLU | score: about 5 points drop | Hurt |
| apply steganographic malicious finetuning | GPT-4.1 · MGSM | score: about 1 point increase | No clear effect |
| add auxiliary base-4 encoding track to multitask finetuning (ablation) | GPT-4.1 · response quality | response quality | Improved |
| filter all steganographic characters from inputs | defense · stegotext attack | attack success | Improved |
| apply token frequency penalty during generation | defense · stegotext attack | attack success | Improved |
Closest papers by meaning
- Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation (ICML 2024 · similarity 0.89)
- Safety Layers in Aligned Large Language Models: The Key to LLM Security (ICLR 2025 · similarity 0.89)
- Soft Prompt Threats: Attacking Safety Alignment and Unlearning in Open-Source LLMs through the Embedding Space (NeurIPS 2024 · similarity 0.87)
- Does Safety Training of LLMs Generalize to Semantically Related Natural Prompts? (ICLR 2025 · similarity 0.87)
- Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy (CVPR 2025 · similarity 0.87)
- Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models? (CVPR 2025 · 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.
