AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Reasoned Safety Alignment: Ensuring Jailbreak Defense via Answer-Then-Check

Chentao Cao, Xiaojun Xu, Bo Han, Hang Li

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Jailbreak attacks and defenses · 6 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 Answer-Then-Check reasoning (ReSA-SFT) vs base modelLlama3.1-8B-Instruct · StrongREJECT jailbreaks avg (LlamaGuard)defense success rate: 0.5008 -> 0.9046Improved
add RL stage after ReSA-SFTLlama3.1-8B-Instruct · StrongREJECT jailbreaks avg (LlamaGuard)defense success rate: 0.9046 -> 0.9932Improved
ReSA-SFT vs WJ-SFT against adaptive PAIR attackLlama3.1-8B-Instruct · StrongREJECT PAIRdefense success rate: 0.3291 -> 0.6965Improved
ReSA-SFT safety tuningLlama3.1-8B-Instruct · MATH500accuracy: 50.60% -> 49.00%Hurt
ReSA-RLQwen2.5-7B-Instruct · MATH500accuracy: 77.00% -> 75.40%Hurt
ReSA-SFT vs base on over-refusalLlama3.1-8B-Instruct · over-refusal average (XSTest, OKTest, WJ-Eval)over-refusal accuracy: 93.27% -> 95.02%Improved
prompt engineering (goal priority) on deepseek-v3deepseek-v3 · XSTestover-refusal accuracy: -13.60%Hurt
train on only 0.5K ReSA samples (ablation)Qwen2.5-7B-Instruct · jailbreak safety averagedefense successImproved
use ReSA data with only WildJailbreak prompts (63K) vs WJ-SFT (262K) (ablation)Llama3.1-8B-Instruct · multiple jailbreak methodsdefense success rateImproved
add diverse jailbreak types to training data (full ReSA vs Only WJ) (ablation)ReSA-SFT · TAP and ReNeLLMdefense success rateImproved
adaptive Answer-Then-Check (skip check on normal queries) (ablation)Qwen2.5-7B-Instruct · MATH500 runtimeruntime: 77s -> 62sImproved
ReSA-SFT on harmful queries runtimeQwen2.5-7B-Instruct · StrongREJECT runtimeruntime: 177s -> 46sImproved

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