Research map / Adversarial robustness and security / Attacks on text and graphs
MirrorShield: Towards Dynamic Adaptive Defense Against Jailbreaks via Entropy-Guided Mirror Crafting
Rui Pu, Chaozhuo Li, Rui Ha, Litian Zhang, Lirong Qiu, Xi Zhang
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 length constraint to mirror generation (ablation) | MirrorShield · BaitAttack (Llama2) | ASR: 0.654 -> 0.009 | Improved |
| add syntax constraint to mirror generation (ablation) | MirrorShield · BaitAttack (Llama2) | ASR: 0.243 -> 0.009 | Improved |
| add sentiment constraint to mirror generation (ablation) | MirrorShield · BaitAttack (Llama2) | ASR: 0.071 -> 0.009 | Improved |
| add multiple-query guidance (ablation) | MirrorShield · BaitAttack (Vicuna) | ASR: 0.042 -> 0.028 | Improved |
| use MirrorShield instead of SafeDecoding | Vicuna-13b-v1.5 · Average ASR over 12 datasets/attacks | ASR: 0.083 -> 0.036 | Improved |
| use MirrorShield on well-aligned model | Llama2-7b-chat · Average ASR over 12 datasets/attacks | ASR: 0.315 -> 0.006 | Improved |
| use MirrorShield instead of no defense | Llama2-7b-chat · AlpacaEval | WinRate: 69.6% -> 68.6% | Hurt |
| use Paraphrase defense | Llama2-7b-chat · AlpacaEval | WinRate: 69.6% -> 34.1% | Hurt |
| raise RIU threshold sigma from 0.80 to 0.90 (ablation) | MirrorShield · 100 benign + 100 jailbreak prompts (Llama2) | detection accuracy: 0.97 -> 0.75 | Hurt |
| increase candidate mirrors to 5 or more (ablation) | MirrorShield · BaitAttack (Llama2) | ASR: below 0.3 at 4 candidates, near zero at 5+ | Improved |
Closest papers by meaning
- Systematic Scaling Analysis of Jailbreak Attacks in Large Language Models (NeurIPS 2026 · similarity 0.89)
- MetaDefense: Defending Fine-tuning based Jailbreak Attack Before and During Generation (NeurIPS 2025 · similarity 0.89)
- h4rm3l: A Language for Composable Jailbreak Attack Synthesis (ICLR 2025 · similarity 0.89)
- One Model Transfer to All: On Robust Jailbreak Prompts Generation against LLMs (ICLR 2025 · similarity 0.88)
- JailPO: A Novel Black-Box Jailbreak Framework via Preference Optimization Against Aligned LLMs (AAAI 2025 · similarity 0.88)
- AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models (ICLR 2024 · similarity 0.88)
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.
