Refers to the practice of placing text, instructions, keywords, or other content in a résumé or job application that is intended to influence an artificial intelligence system used to screen, evaluate, or rank applicants. The inserted content may be hidden or difficult for a human reviewer to see. This works by AI introducing very small text, text that matches the background color, or other formatting, which can still be read by an AI system processing the document. A prompt might attempt to tell the system to treat the applicant as highly qualified, increase the applicant's ranking, or otherwise influence the screening process without adding legitimate evidence of the person's qualifications.

Résumé prompt injection is just one a form of prompt injection, a broader AI and cybersecurity problem in which instructions embedded in information processed by an AI system can cause the system to behave differently from what its developers or users intended. Because the instructions reach the AI system through an external document rather than being entered directly by the system's user, the practice may also be described technically as indirect prompt injection.

Research published in 2026 provides evidence that the practice is occurring in actual hiring systems. A Duke University-led study examining approximately 200,000 real résumés found prompt injections in about 1% of the résumés studied and reported that their prevalence had increased substantially in recent years.

Researchers caution that the presence of hidden material does not necessarily demonstrate that every applicant intentionally attempted to manipulate a hiring system; some individuals may use résumé templates containing hidden text without realizing it.

The practice raises questions about the reliability and security of AI assisted hiring, fairness among applicants, the integrity of résumé screening, and the continuing need for human oversight of automated hiring systems.

See Topic Brief: Emerging Issues related to Human-Readable & Machine-Interpreted Information | Learn & Work Ecosystem Library