New Research Uses Pre-Decoding States in Diffusion LLMs for Jailbreak Detection

An arXiv paper proposes detecting jailbreak attempts in diffusion large language models by leveraging their pre-decoding internal states, positioning dLLMs as a strong alternative to autoregressive models.

Read the full story at arXiv cs.CL ↗

Timeline · 1 report

1. 2026-11-10 05:00 · arXiv cs.CL
   Beyond the Prompt: Leveraging Pre-Decoding States for Jailbreak Detection in dLLMs