List your company

Safety alignment against harmful fine-tuning on BeaverTails

60.9Harm Score (HS)Vaccine
6.320.47534.6548.825May 23, 2026

Evaluation results

MethodLinks
VaccineBackbone=LLaMA3-8B2026.0560.927.4
BoosterBackbone=LLaMA3-8B2026.0558.140
RepNoiseBackbone=LLaMA3-8B2026.0557.237.9
Buffer-and-ReinforceBackbone=LLaMA3-8B2026.058.477.5