Prompt Injection in Automated Résumé Screening with Large Language Models: Single and Multi-Injection Settings

Researchers have discovered a vulnerability in automated résumé screening systems that utilize large language models (LLMs), where subtle self-promotional text, known as prompt injection, can manipulate the algorithm's evaluation. This technique introduces no new qualifications but is designed to influence the LLM's ranking of job applicants. Through controlled experiments, it has been shown that prompt injection can significantly impact the outcome of automated résumé screening, allowing candidates to strategically manipulate the system¹. The use of LLMs in hiring systems creates an incentive for candidates to exploit such vulnerabilities, potentially leading to unfair advantages. The security implications of LLM developments, particularly in high-stakes applications like hiring, are significant and warrant further investigation. This vulnerability matters to practitioners because it highlights the need for robust testing and validation of AI-powered hiring systems to prevent manipulation and ensure fairness in the hiring process.