Twin Agent: Context Residual Compression for Privilege Separated Agents

Abstract

Large language model (LLM) agents are vulnerable to security risks, such as prompt injection attacks from untrusted context that manipulate downstream reasoning and tool use. Existing secure-by-design approaches mitigate this risk by separating untrusted observations from privileged execution and careful control of information flow, but often degrade utility and require extensive task-specific engineering. We thus propose Twin Agent, a general privilege separation design pattern inspired by resi

Categories

Framework mappings

Suggested from the entry's categories.

@misc{hu2026twin,
  title = {{Twin Agent: Context Residual Compression for Privilege Separated Agents}},
  author = {Zhanhao Hu and Dennis Jacob and Xiao Huang and Zhaorun Chen and Bo Li and David Wagner},
  year = {2026},
  month = jul,
  eprint = {2607.19595},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.19595}
}