Probing Latent Subspaces in LLM for AI Security: Identifying and Manipulating Adversarial States

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities across various tasks, yet they remain vulnerable to adversarial manipulations such as jailbreaking via prompt injection attacks. These attacks bypass safety mechanisms to generate restricted or harmful content. In this study, we investigated the underlying latent subspaces of safe and jailbroken states by extracting hidden activations from a LLM. Inspired by attractor dynamics in neuroscience, we hypothesized that LLM activat

Categories

Framework mappings

Suggested from the entry's categories.

@misc{chia2025probing,
  title = {{Probing Latent Subspaces in LLM for AI Security: Identifying and Manipulating Adversarial States}},
  author = {Xin Wei Chia and Jonathan Pan},
  year = {2025},
  month = mar,
  eprint = {2503.09066},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2503.09066},
  url = {https://www.semanticscholar.org/paper/03f458d740dea240d07abe93cd7e756cb6a20cb8}
}