WildJailbreak

The Allen Institute for AI introduced BenchMIRT, a new method for auditing LLM benchmarks at the level of individual prompts, which uses multidimensional Item Response Theory to separate the underlyin…

A reproducibility study of latent-space safety probes found that lightweight MLP probes trained on final-layer activations of LLaMA-3.1-8B detect harmful prompts with F1 scores within 0.37 percentage …

Researchers propose Dialogue Critic Guided Sampling (DCGS), a framework that infers user intent at every turn of dialogue to defend large language models against multi-turn attacks. DCGS models advers…