SafeDream: Safety World Model for Proactive Early Jailbreak Detection

Abstract

Multi-turn jailbreak attacks progressively erode LLM safety alignment across seemingly innocuous conversation turns, achieving success rates exceeding 90% against state-of-the-art models. Existing alignment-based and guardrail methods suffer from three key limitations: they require costly weight modification, evaluate each turn independently without modeling cumulative safety erosion, and detect attacks only after harmful content has been generated. To address these limitations, we first formula

Categories

Framework mappings

Suggested from the entry's categories.

@misc{yan2026safedream,
  title = {{SafeDream: Safety World Model for Proactive Early Jailbreak Detection}},
  author = {Bo Yan and Weikai Lin and Yada Zhu and Song Wang},
  year = {2026},
  month = apr,
  eprint = {2604.16824},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2604.16824}
}