Indirect Prompt Injection CV Dataset
Milani, Alfredo · MILANI, Alfredo · Franzoni, Valentina · et al.
15 KB
This dataset has been used for the research described in the work: Alfredo Milani, Valentina Franzoni, Emanuele Florindi, Indirect Prompt Injection in Large Language Models, (2026) , Neural Computing and Application, Springer to evaluate the robustness of LLMs against indirect prompt injection, we designed a systematic experimental framework based on a heterogeneous dataset of Curriculum Vitae (CV) documents. The scope of the dataset is to verify the generalizability of threats and countermeasures. We designed a controlled matrix of 27 unique CVs to isolate the impact of document length and content density on model vulnerability.