AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

AlignTree: Efficient Defense Against LLM Jailbreak Attacks

Gil Goren, Shahar Katz, Lior Wolf

AAAI 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 3 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
add RBF-SVM non-linear signals (vs linear SVMs) (ablation)AlignTree · MalwareGenASR (Qwen2.5-0.5B): 61.0 -> 4.0Improved
add RBF-SVM non-linear signals (vs linear SVMs) (ablation)AlignTree · MalwareGenASR (Gemma-3-12b): 8.0 -> 10.0Hurt
add SVM signals to single refusal direction (ablation)AlignTree · MalwareGenASR (Qwen2.5-0.5B): 89.0 -> 4.0Improved
add refusal-direction activations to SVM signals (ablation)AlignTree · MalwareGenASR (Qwen2.5-0.5B): 33.0 -> 4.0Improved
use multiple refusal vectors instead of one (ablation)RefusalClassifier · MalwareGenASR (Qwen2.5-0.5B): 89.0 -> 29.0Improved
AlignTree defense vs no defenseQwen2.5-0.5B-Instruct · MalwareGenASR: 91.0 -> 4.0Improved
AlignTree defense vs no defenseGemma-3-12b-it · PAIRASR: 36.0 -> 10.0Improved
AlignTree vs AutoDefense on PromptInjectQwen2.5-0.5B-Instruct · PromptInjectASR: 0 -> 41.0Hurt
AlignTree vs SelfDefense on benign refusalLlama-3.1-8B-Instruct · PIQArefusal rate: 51.0 -> 1.0Improved
AlignTree vs PerplexityDefense on ASRQwen2.5-0.5B-Instruct · MalwareGenASR: 84.0 -> 4.0Improved
AlignTree vs no defense on LlamaLlama-3.1-8B-Instruct · MalwareGenASR: 9.0 -> 5.0No clear effect

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