CPInj: Uncovering Prompt Injection Risks in Textual Collaborative Prompt Optimization

Abstract

Textual Collaborative Prompt Optimization (TCPO) extends Textgrad (Yuksekgonul et al., 2025) to a decentralized setting by allowing multiple clients to jointly improve prompts for large language models (LLMs) while keeping their data locally. Its reliance on free-form textual updating and aggregation introduces a new and largely unexplored attack surface, i.e., malicious instructions can be injected into local prompts and propagated through server-side prompt aggregation. Unlike conventional pro

Categories

Framework mappings

Suggested from the entry's categories.

@misc{liao2026cpinj,
  title = {{CPInj: Uncovering Prompt Injection Risks in Textual Collaborative Prompt Optimization}},
  author = {Xinting Liao and Behnoosh Zamanlooy and Masoumeh Shafieinejad and David B. Emerson and Ruinan Jin and Deval Pandya and Xiaoxiao Li},
  year = {2026},
  month = jul,
  eprint = {2607.18622},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.18622}
}