Research map / Large language models / Instruction and fine-tuning
Reflector: Internalizing Step-wise Reflection against Indirect Jailbreaks
Jiachen Ma, Jiawen Zhang, Xiangtian Li, Bo Zou, Chaochao Lu, Chao Yang
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 SFT stage before RL (two-stage) (ablation) | Reflector · DRA | DSR: 87.88% -> 92.31% | Improved |
| add SFT stage before RL (two-stage) (ablation) | Reflector · AdvGLUE | accuracy: 54.20% -> 68.29% | Improved |
| add GDPO RL after SFT | Reflector · DRA | DSR: 88.16% -> 92.31% | Improved |
| add reflection bonus (lambda 0.0 -> 0.3) (ablation) | Reflector · DRA | DSR: 83.65% -> 92.31% | Improved |
| add reflection bonus (lambda 0.0 -> 0.3) (ablation) | Reflector · AdvGLUE | accuracy: 59.62% -> 68.29% | Improved |
| increase lambda from 0.3 to 0.8 (ablation) | Reflector · MMLU-Pro | accuracy: 45.20% -> 42.15% | Hurt |
| increase lambda from 0.3 to 0.8 (ablation) | Reflector · AdvGLUE | accuracy: 68.29% -> 60.03% | Hurt |
| suppress reflect token at decoding (ablation) | Reflector · DRA | DSR: 92.31% -> 90.76% | No clear effect |
| change indirect attack source for training data (ablation) | Reflector · DRA | DSR: 90.58% - 92.31% | No clear effect |
| replace GPT-5 teacher with GPT-3.5-turbo (ablation) | Reflector · DRA | DSR: 92.31% -> 90.85% | No clear effect |
| use Qwen-Guard as weaker reward model (ablation) | Reflector · DRA | DSR: 89.40% | No clear effect |
| raise safety-to-general data ratio too high (ablation) | Reflector · MMLU | accuracy | Hurt |
Closest papers by meaning
- Metis: Learning to Jailbreak LLMs via Self-Evolving Metacognitive Policy Optimization (ICML 2026 · similarity 0.87)
- Systematic Scaling Analysis of Jailbreak Attacks in Large Language Models (NeurIPS 2026 · similarity 0.86)
- MetaCipher: A Time-Persistent and Universal Multi-Agent Framework for Cipher-Based Jailbreak Attacks for LLMs (AAAI 2026 · similarity 0.86)
- AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models (ICLR 2024 · similarity 0.85)
- Lifelong Safety Alignment for Language Models (NeurIPS 2025 · similarity 0.85)
- One Model Transfer to All: On Robust Jailbreak Prompts Generation against LLMs (ICLR 2025 · similarity 0.85)
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.
