WildGuard

Researchers introduced ProbGuard in an August 11 preprint as a guardrail that estimates the probability an unfinished language-model response will become unsafe, using output distributions from as few…

A prompt injection classifier trained on a 14M-parameter Electra-small discriminator achieves 0.990 F1 and 0.999 ROC-AUC on English WildGuardTest prompt-adversarial classification, but shows edge case…

DeshwalX released electra-small-prompt-injection-v1, a 14M-parameter multi-label classifier based on google/electra-small-discriminator, achieving 90% Micro F1 and 89% Macro F1 on the WildGuard test b…