AlcaTRAz - Anchored Tree-Rule Defense Against Jailbreaks

Abstract

Large language models (LLMs) are vulnerable to jailbreak attacks that bypass safety alignment through carefully crafted prompts. Many existing defenses require access to model weights or internals, making them difficult to apply to black-box deployments. We propose AlcaTRAz (Anchored Tree-Rule defense Against jailbreaks), a prompt-level defense based on rule trees that operates exclusively on the input text and requires no modification or retraining of the target model. The method automatically

Categories

Framework mappings

Suggested from the entry's categories.

@misc{res2026alcatraz,
  title = {{AlcaTRAz - Anchored Tree-Rule Defense Against Jailbreaks}},
  author = {Jakub Reš and Petr Kaška and Martin Perešíni and Martin Ukrop and Kamil Malinka},
  year = {2026},
  month = sep,
  eprint = {2609.03693},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.03693}
}