List your company

Safety Refusal on AdvBench

99.42Refusal RateLow-Rank Combination
6.631230.720654.8178.8994May 22, 2025Jul 10, 2025Aug 28, 2025Oct 16, 2025Dec 4, 2025Jan 22, 2026Mar 12, 2026

Evaluation results

MethodLinks
Low-Rank CombinationBackbone=REFUSE-LLAMA, Method=Low-Rank Combination2026.0399.42
Categorical SteeringBackbone=REFUSE-LLAMA, Method=Categorical Steering2026.0399.23
WASBackbone=Llama 3.1-8B2025.0598.8
WASBackbone=Llama 3.2-1B2025.0598.2
WASBackbone=Mistral 7B2025.0598.2
CASTBackbone=Llama 3.1-8B2025.0595.7
Low-Rank CombinationBackbone=DEEPSEEK R1 DISTILL LLAMA, Method=Low-Rank Combination2026.0394.81
REFUSE-LLAMABackbone=REFUSE-LLAMA2026.0394.23
Low-Rank CombinationBackbone=LLAMA 3 8B INSTRUCT, Method=Low-Rank Combination2026.0394.23
DEEPSEEK R1 DISTILL LLAMABackbone=DEEPSEEK R1 DISTILL LLAMA2026.0394.23
LLAMA 3 8B INSTRUCTBackbone=LLAMA 3 8B INSTRUCT2026.0393.85
CASTBackbone=Llama 3.2-1B2025.0593.7
CASTBackbone=Mistral 7B2025.0591.7
Base ModelBackbone=Llama 3.1-8B2025.0591.5
Base ModelBackbone=Llama 3.2-1B2025.0591.1
ITIBackbone=Llama 3.1-8B2025.0589.1
ACTBackbone=Llama 3.1-8B2025.0583.9
ACTBackbone=Llama 3.2-1B2025.0583.1
ITIBackbone=Llama 3.2-1B2025.0582.5
GER-SteerBackbone=Qwen2.5-7B2026.0377.5
GER-SteerBackbone=Gemma-2-9B-it2026.0377.3
ACTBackbone=Gemma-2-9B-it2026.0377
LDPBackbone=Gemma-2-9B-it2026.0376.7
Angular SteeringBackbone=Gemma-2-9B-it2026.0376.5
NL-ITIBackbone=Gemma-2-9B-it2026.0376.2
NL-ITIBackbone=Qwen2.5-7B2026.0375.3
RePEBackbone=Gemma-2-9B-it2026.0375.2
CAABackbone=Qwen2.5-7B2026.0375.1
CAABackbone=Gemma-2-9B-it2026.0375.1
Angular SteeringBackbone=Qwen2.5-7B2026.0374.7
VanillaBackbone=Gemma-2-9B-it2026.0374.5
RePEBackbone=Qwen2.5-7B2026.0374.2
ACTBackbone=Qwen2.5-7B2026.0373.2
GER-SteerBackbone=Llama-3.1-8B-Instruct2026.0367.7
RePEBackbone=Llama-3.1-8B-Instruct2026.0367.5
Angular SteeringBackbone=Llama-3.1-8B-Instruct2026.0367.4
LDPBackbone=Llama-3.1-8B-Instruct2026.0367.1
ACTBackbone=Llama-3.1-8B-Instruct2026.0366.8
CAABackbone=Llama-3.1-8B-Instruct2026.0366.7
NL-ITIBackbone=Llama-3.1-8B-Instruct2026.0366.4
LDPBackbone=Qwen2.5-7B2026.0366.3
VanillaBackbone=Llama-3.1-8B-Instruct2026.0365.2
VanillaBackbone=Qwen2.5-7B2026.0364.7
ITIBackbone=Mistral 7B2025.0554.9
ACTBackbone=Mistral 7B2025.0551.4
Base ModelBackbone=Mistral 7B2025.0510.2