List your company

Backdoor Defense on BadEdit, VPI, BadChain, and EmbedX

93.5Utility AccuracyNo Defense
77.79681.87385.9590.027Apr 27, 2026

Evaluation results

MethodLinks
No DefenseModel=DS-R1 Distill-70B, Architecture=Reasoning-Focused Models2026.0493.598.897.899.699.2
DemoDefenseModel=DS-R1 Distill-70B, Architecture=Reasoning-Focused Models2026.0492.213.5442.830.5
TIGSModel=DS-R1 Distill-70B, Architecture=Reasoning-Focused Models2026.0491.812.511.220.823.1
No DefenseModel=DS-R1 Distill-32B, Architecture=Reasoning-Focused Models2026.0491.298.597.299.498.8
W2SDefenseModel=DS-R1 Distill-70B, Architecture=Reasoning-Focused Models2026.0490.25.88.222.816.5
DemoDefenseModel=DS-R1 Distill-32B, Architecture=Reasoning-Focused Models2026.0489.515.24.540.528.2
TIGSModel=DS-R1 Distill-32B, Architecture=Reasoning-Focused Models2026.0489.113.210.821.522.4
No DefenseModel=DS-R1 Distill-7B, Architecture=Reasoning-Focused Models2026.0488.697.896.599.198.6
No DefenseModel=Qwen3-MoE 80B-A3B, Architecture=Sparse Mixture-of-Experts2026.0488.296.594.895.598.4
DemoDefenseModel=Qwen3-MoE 80B-A3B, Architecture=Sparse Mixture-of-Experts2026.0487.517.55.524.818.5
W2SDefenseModel=DS-R1 Distill-32B, Architecture=Reasoning-Focused Models2026.0487.25.27.620.515.2
TIGSModel=Qwen3-MoE 80B-A3B, Architecture=Sparse Mixture-of-Experts2026.0487.27.86.813.617.4
No DefenseModel=Llama-3-70B, Architecture=Dense Transformers2026.0486.599.596.898.898.2
DemoDefenseModel=DS-R1 Distill-7B, Architecture=Reasoning-Focused Models2026.0486.417.55.238.225.6
TIGSModel=Llama-3-70B, Architecture=Dense Transformers2026.048611.85.512.815.6
DemoDefenseModel=Llama-3-70B, Architecture=Dense Transformers2026.0485.812.52.825.415.5
W2SDefenseModel=Qwen3-MoE 80B-A3B, Architecture=Sparse Mixture-of-Experts2026.0485.26.27.216.518.2
No DefenseModel=Qwen3-MoE 30B-A3B, Architecture=Sparse Mixture-of-Experts2026.0485.195.493.194.297.6
W2SDefenseModel=Llama-3-70B, Architecture=Dense Transformers2026.04854.8613.812.4
TIGSModel=DS-R1 Distill-7B, Architecture=Reasoning-Focused Models2026.0484.811.49.519.820.5
W2SDefenseModel=DS-R1 Distill-7B, Architecture=Reasoning-Focused Models2026.0484.14.56.818.413.5
No DefenseModel=Llama-3-8B, Architecture=Dense Transformers2026.0482.599.295.898.497.7
DemoDefenseModel=Qwen3-MoE 30B-A3B, Architecture=Sparse Mixture-of-Experts2026.0482.319.56.826.219.4
TIGSModel=Qwen3-MoE 30B-A3B, Architecture=Sparse Mixture-of-Experts2026.0481.65.46.111.515.2
DemoDefenseModel=Llama-3-8B, Architecture=Dense Transformers2026.0481.516.33.728.218
TIGSModel=Llama-3-8B, Architecture=Dense Transformers2026.0480.59.64.210.514
W2SDefenseModel=Llama-3-8B, Architecture=Dense Transformers2026.0480.45.24.811.611.6
W2SDefenseModel=Qwen3-MoE 30B-A3B, Architecture=Sparse Mixture-of-Experts2026.0478.45.66.414.816.5