List your company

Privacy Flow Control on CONFAIDE Tier 3 InfoFlow-Control

50LeakageMistral-7B (Drunk language inducement)
4861.57588.5Jan 19, 2026

Evaluation results

MethodLinks
Mistral-7B (Drunk language inducement)Variant=RL2026.0150337780
Mistral-7B (Drunk language inducement)Variant=FT2026.0157106678
Mistral-7BVariant=Base2026.0169107757
GPT-3.5Variant=Base2026.018006546
Mistral-7B (Drunk language inducement)Variant=Pr.2026.0181787767
GPT-4 (Drunk language inducement)Variant=FT2026.0196759997
GPT-4 (Drunk language inducement)Variant=Pr.2026.0197549080
LLaMA2-7B (Drunk language inducement)Variant=Pr.2026.0197100100100
LLaMA2-7B (Drunk language inducement)Variant=RL2026.0197100100100
GPT-3.5 (Drunk language inducement)Variant=Pr.2026.019873100100
LLaMA2-7BVariant=Base2026.019810010099
LLaMA2-7B (Drunk language inducement)Variant=FT2026.0198100100100
LLaMA3-8B (Drunk language inducement)Variant=RL2026.0198100100100
GPT-4Variant=Base2026.011006127
LLaMA3-8BVariant=Base2026.01100100100100
LLaMA3-8B (Drunk language inducement)Variant=Pr.2026.0110082100100
LLaMA3-8B (Drunk language inducement)Variant=FT2026.01100100100100