Prompt injection attacks on LLM hiring systems are effective only when rare and candidate quality is homogeneous; widespread adoption or quality differences make the attack ineffective, but fairness concerns remain when manipulation is selective.
This paper studies how candidates can manipulate LLM-based résumé screening systems through prompt injection—adding subtle persuasive text without new qualifications. Experiments show the attack works when few candidates use it and qualifications are similar, but fails as manipulation spreads. The research reveals fairness risks when lower-quality candidates can game the system.