●researcherYou can study the vulnerability of model evaluation protocols to extraction attacks.
●founderYour proprietary evaluation datasets and judge models may be vulnerable to low-cost extraction attacks.
●policyThis reveals a significant security risk regarding the theft of proprietary evaluation logic.