AI Paper Atlas

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

Safety Instincts: LLMs Learn to Trust Their Internal Compass for Self-Defense

Guobin Shen, Dongcheng Zhao, Haibo Tong, Jindong Li, Feifei Zhao, Yi Zeng

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 2 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 entropy-minimizing reward (SIRL) instead of no safety trainingLlama-3.1-8B-Instruct · JailbreakBench (DSR)defense success rate (%): 84.3 -> 99.1Improved
add SIRLQwen2.5-7B-Instruct · JailbreakBench (DSR)defense success rate (%): 82.8 -> 99.9Improved
add SIRLLlama-3.2-3B-Instruct · JailbreakBench (DSR)defense success rate (%): 95.6 -> 100.0Improved
use SIRL instead of SFTLlama-3.1-8B-Instruct · AlpacaEvalwin rate: 19.1 -> 51.9Improved
use SIRL instead of SFTQwen2.5-7B-Instruct · MATH-500accuracy (%): 31.8 -> 78.6Improved
add SIRLLlama-3.1-8B-Instruct · MATH-500accuracy (%): 49.0 -> 51.2Improved
add SIRLLlama-3.2-3B-Instruct · MATH-500accuracy (%): 42.2 -> 41.4Hurt
replace min-entropy reward with maximizing entropy (neg-SIRL) (ablation)SIRL · JailbreakBench (DSR)defense success rate (%): 99.1 -> 72.1Hurt
add entropy reward instead of random reward (ablation)SIRL · JailbreakBench (DSR)defense success rate (%): 85.2 -> 99.1Improved
replace entropy reward with minimizing perplexity (ablation)SIRL · JailbreakBench (DSR)defense success rate (%): 99.1 -> 98.7No clear effect
lower KL coefficient beta to 1e-4 (ablation)SIRL · JailbreakBench (DSR)defense success rate (%): 99.7% DSR; BBH drops 5.6 points and HumanEval 3.7 pointsMixed
raise KL coefficient beta to 1e-2 (ablation)SIRL · JailbreakBench (DSR)defense success rate (%): 93.6% DSRHurt

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