Measuring Real-World Prompt Injection Attacks in LLM-based Resume Screening

A systematic study has been conducted on prompt injection attacks in LLM-based resume screening, revealing that approximately 1% of real-world resumes contain such vulnerabilities. This research, based on a dataset of around 200,000 resumes collected by hireEZ, demonstrates tailored detection methods that outperform existing general-purpose detectors.

The findings highlight a significant security concern in the use of LLMs for resume screening, suggesting that organizations relying on these technologies must enhance their defenses against prompt injection attacks to ensure the integrity of their hiring processes.