Multi-turn Jailbreak Defense
89.2Success RateMASCing
Evaluation Results
| Method | Links | |
|---|---|---|
| MASCingModel=Qwen3-30B-A3B-Instruct-2507, Steering Strategy=Activation Steering2026.04 | 89.2 | |
| MASCingModel=GPT-OSS-20B, Steering Strategy=Activation Steering2026.04 | 87.9 | |
| MASCingModel=Qwen1.5-MoE-A2.7B-Chat, Steering Strategy=Activation Steering2026.04 | 87.2 | |
| MASCingModel=DeepSeek-MoE-16B-Chat, Steering Strategy=Activation Steering2026.04 | 84.9 | |
| MASCingModel=Average, Steering Strategy=Activation Steering2026.04 | 83.9 | |
| MASCingModel=Phi-3.5-MoE-Instruct, Steering Strategy=Activation Steering2026.04 | 80.6 | |
| MASCingModel=Hunyuan-A13B-Instruct, Steering Strategy=Activation Steering2026.04 | 80.4 | |
| MASCingModel=Mixtral-8x7B-Instruct-v0.1, Steering Strategy=Activation Steering2026.04 | 77.1 | |
| Expert SteeringModel=GPT-OSS-20B, Steering Strategy=Expert Steering2026.04 | 74.2 | |
| Expert SteeringModel=Qwen3-30B-A3B-Instruct-2507, Steering Strategy=Expert Steering2026.04 | 73.5 | |
| Expert SteeringModel=Qwen1.5-MoE-A2.7B-Chat, Steering Strategy=Expert Steering2026.04 | 70.1 | |
| Expert SteeringModel=DeepSeek-MoE-16B-Chat, Steering Strategy=Expert Steering2026.04 | 69.1 | |
| Expert SteeringModel=Average, Steering Strategy=Expert Steering2026.04 | 69 | |
| Expert SteeringModel=Phi-3.5-MoE-Instruct, Steering Strategy=Expert Steering2026.04 | 68.7 | |
| SteerMoEModel=Mixtral-8x7B-Instruct-v0.12026.04 | 66.8 | |
| Expert SteeringModel=Hunyuan-A13B-Instruct, Steering Strategy=Expert Steering2026.04 | 65.3 | |
| BaselineModel=GPT-OSS-20B2026.04 | 61.8 | |
| Expert SteeringModel=Mixtral-8x7B-Instruct-v0.1, Steering Strategy=Expert Steering2026.04 | 61.8 | |
| SteerMoEModel=GPT-OSS-20B2026.04 | 61.3 | |
| SteerMoEModel=DeepSeek-MoE-16B-Chat2026.04 | 59.7 | |
| SteerMoEModel=Phi-3.5-MoE-Instruct2026.04 | 59 | |
| SteerMoEModel=Average2026.04 | 58.4 | |
| SteerMoEModel=Qwen1.5-MoE-A2.7B-Chat2026.04 | 57.9 | |
| BaselineModel=Phi-3.5-MoE-Instruct2026.04 | 57.7 | |
| BaselineModel=DeepSeek-MoE-16B-Chat2026.04 | 52.9 | |
| SteerMoEModel=Qwen3-30B-A3B-Instruct-25072026.04 | 52.6 | |
| BaselineModel=Average2026.04 | 52.5 | |
| BaselineModel=Qwen1.5-MoE-A2.7B-Chat2026.04 | 51.6 | |
| SteerMoEModel=Hunyuan-A13B-Instruct2026.04 | 51.4 | |
| BaselineModel=Hunyuan-A13B-Instruct2026.04 | 48.8 | |
| BaselineModel=Mixtral-8x7B-Instruct-v0.12026.04 | 47.7 | |
| BaselineModel=Qwen3-30B-A3B-Instruct-25072026.04 | 47.3 |