List your company

Backdoor Detection on LMSYS-Chat-1M

0.1FPR on LMSYSbadnet-Z-step300
0.0520.3760.71.024Jul 31, 2025

Evaluation results

MethodLinks
badnet-Z-step300Training Method=FFT, Attack Type=badnet, Target Identifier=Z, Training Steps=3002025.070.1
ctba-Y-step100Training Method=FFT, Attack Type=ctba, Target Identifier=Y, Training Steps=1002025.070.1
mtba-Y-step100Training Method=FFT, Attack Type=mtba, Target Identifier=Y, Training Steps=1002025.070.1
mtba-Y-step500Training Method=FFT, Attack Type=mtba, Target Identifier=Y, Training Steps=5002025.070.1
mtba-Z-step500Training Method=FFT, Attack Type=mtba, Target Identifier=Z, Training Steps=5002025.070.1
badnet-Y-step300Training Method=FFT, Attack Type=badnet, Target Identifier=Y, Training Steps=3002025.070.2
ctba-X-step100Training Method=FFT, Attack Type=ctba, Target Identifier=X, Training Steps=1002025.070.2
ctba-Y-step300Training Method=FFT, Attack Type=ctba, Target Identifier=Y, Training Steps=3002025.070.2
mtba-X-step100Training Method=FFT, Attack Type=mtba, Target Identifier=X, Training Steps=1002025.070.2
mtba-X-step300Training Method=FFT, Attack Type=mtba, Target Identifier=X, Training Steps=3002025.070.2
mtba-Z-step100Training Method=FFT, Attack Type=mtba, Target Identifier=Z, Training Steps=1002025.070.2
badnet-X-step100Training Method=FFT, Attack Type=badnet, Target Identifier=X, Training Steps=1002025.070.3
badnet-X-step500Training Method=FFT, Attack Type=badnet, Target Identifier=X, Training Steps=5002025.070.3
badnet-Y-step500Training Method=FFT, Attack Type=badnet, Target Identifier=Y, Training Steps=5002025.070.3
badnet-Z-step100Training Method=FFT, Attack Type=badnet, Target Identifier=Z, Training Steps=1002025.070.3
ctba-X-step300Training Method=FFT, Attack Type=ctba, Target Identifier=X, Training Steps=3002025.070.3
ctba-Y-step500Training Method=FFT, Attack Type=ctba, Target Identifier=Y, Training Steps=5002025.070.3
ctba-Z-step300Training Method=FFT, Attack Type=ctba, Target Identifier=Z, Training Steps=3002025.070.3
ctba-Z-step500Training Method=FFT, Attack Type=ctba, Target Identifier=Z, Training Steps=5002025.070.3
mtba-X-step500Training Method=FFT, Attack Type=mtba, Target Identifier=X, Training Steps=5002025.070.3
mtba-Y-step300Training Method=FFT, Attack Type=mtba, Target Identifier=Y, Training Steps=3002025.070.3
badnet-X-step300Training Method=FFT, Attack Type=badnet, Target Identifier=X, Training Steps=3002025.070.4
badnet-Y-step100Training Method=FFT, Attack Type=badnet, Target Identifier=Y, Training Steps=1002025.070.4
badnet-Y-step400Training Method=FFT, Attack Type=badnet, Target Identifier=Y, Training Steps=4002025.070.4
badnet-Z-step500Training Method=FFT, Attack Type=badnet, Target Identifier=Z, Training Steps=5002025.070.4
ctba-Z-step100Training Method=FFT, Attack Type=ctba, Target Identifier=Z, Training Steps=1002025.070.4
mtba-Z-step200Training Method=FFT, Attack Type=mtba, Target Identifier=Z, Training Steps=2002025.070.4
mtba-Z-step300Training Method=FFT, Attack Type=mtba, Target Identifier=Z, Training Steps=3002025.070.4
mtba-Z-step400Training Method=FFT, Attack Type=mtba, Target Identifier=Z, Training Steps=4002025.070.4
badnet-Z-step200Training Method=FFT, Attack Type=badnet, Target Identifier=Z, Training Steps=2002025.070.5
ctba-X-step200Training Method=FFT, Attack Type=ctba, Target Identifier=X, Training Steps=2002025.070.5
ctba-X-step500Training Method=FFT, Attack Type=ctba, Target Identifier=X, Training Steps=5002025.070.5
mtba-X-step200Training Method=FFT, Attack Type=mtba, Target Identifier=X, Training Steps=2002025.070.5
mtba-X-step400Training Method=FFT, Attack Type=mtba, Target Identifier=X, Training Steps=4002025.070.5
mtba-Y-step400Training Method=FFT, Attack Type=mtba, Target Identifier=Y, Training Steps=4002025.070.5
badnet-X-step200Training Method=FFT, Attack Type=badnet, Target Identifier=X, Training Steps=2002025.070.6
badnet-X-step400Training Method=FFT, Attack Type=badnet, Target Identifier=X, Training Steps=4002025.070.6
ctba-X-step400Training Method=FFT, Attack Type=ctba, Target Identifier=X, Training Steps=4002025.070.6
ctba-Y-step400Training Method=FFT, Attack Type=ctba, Target Identifier=Y, Training Steps=4002025.070.6
ctba-Z-step200Training Method=FFT, Attack Type=ctba, Target Identifier=Z, Training Steps=2002025.070.6
ctba-Z-step400Training Method=FFT, Attack Type=ctba, Target Identifier=Z, Training Steps=4002025.070.6
mtba-Y-step200Training Method=FFT, Attack Type=mtba, Target Identifier=Y, Training Steps=2002025.070.6
badnet-Y-step200Training Method=FFT, Attack Type=badnet, Target Identifier=Y, Training Steps=2002025.070.7
badnet-Z-step400Training Method=FFT, Attack Type=badnet, Target Identifier=Z, Training Steps=4002025.070.7
trojan1Training Method=PPO, Attack Type=trojan, Target Identifier=12025.070.7
trojan5Training Method=PPO, Attack Type=trojan, Target Identifier=52025.070.9
ctba-Y-step200Training Method=FFT, Attack Type=ctba, Target Identifier=Y, Training Steps=2002025.071
trojan4Training Method=PPO, Attack Type=trojan, Target Identifier=42025.071
trojan3Training Method=PPO, Attack Type=trojan, Target Identifier=32025.071.1
trojan2Training Method=PPO, Attack Type=trojan, Target Identifier=22025.071.3