List your company

Trigger inversion on Yahoo (test)

30Trigger Inversion Success RateCSO(discrete)
-1.26.91523.1Jun 30, 2026

Evaluation results

MethodLinks
CSO(discrete)Model=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.0630
CSO(discrete)Model=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.0627
ALTModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.0618
ALTModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.0617
MMModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.0615
MMModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.0611
CSO(discrete)Model=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=clean label2026.0610
CSO(discrete)Model=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=dirty label2026.0610
MMModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=clean label2026.0610
MMModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=dirty label2026.0610
ALTModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=clean label2026.0610
ALTModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=dirty label2026.0610
DBSModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=clean label2026.065
DBSModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.063
DBSModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=dirty label2026.062
UATModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.062
UATModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.062
GBDAModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.061
DBSModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.060
UATModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=clean label2026.060
UATModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=dirty label2026.060
PICCOLOModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.060
PICCOLOModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=dirty label2026.060
GBDAModel=Flan-t5-large, Poison rate=high, Trigger type=Tell me seriously(3 tokens), Label type=clean label2026.060
GBDAModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=clean label2026.060
GBDAModel=Flan-t5-large, Poison rate=high, Trigger type=mb(1 token), Label type=dirty label2026.060