Prompt Injection in Automated R\'esum\'e Screening with Large Language Models: Single and Multi-Injection Settings
arXiv:2606.27287v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to screen and rank job applicants, creating incentives for candidates to strategically manipulate algorithmic hiring systems. We study prompt injection in automated r\'esum\'e screening, defined as subtle self-promotional text that introduces no new qualifications but is designed to influence LLM evaluations. Using controlled experiments, we show that prompt injection reliably improves applicant rankings when r\'esum\'e quality is homogeneous and few candidates inject. However, its effectiveness
The increasing adoption of LLMs in critical applications like hiring creates immediate incentives for individuals to manipulate these systems, leading to the rapid emergence of prompt injection techniques.
This highlights a critical vulnerability in autonomous white-collar workflow systems, where subtle manipulations can significantly alter outcomes, requiring robust countermeasures for reliable AI adoption.
LLM-driven screening processes are no longer trustworthy without advanced prompt injection detection, forcing developers and businesses to integrate new security paradigms.
- · AI security firms
- · Ethical AI researchers
- · Companies with advanced AI validation teams
- · Companies relying on unhardened LLM screening
- · Candidates who do not use prompt injection
- · HR departments without AI expertise
Companies will need to invest in more sophisticated prompt engineering and validation techniques to prevent manipulation.
A 'red queen' effect could emerge, where AI manipulation and detection methods continuously evolve, increasing complexity and cost.
Public trust in AI-driven decision-making systems could erode if prompt injection becomes widespread and unaddressed, leading to regulatory interventions.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI