Polishing a résumé with AI is fair game, but hiding instructions to trick an AI screener is deception.

This explores where AI help on a job application stops being legitimate and starts being deceptive. The corpus has no rulebook for this. What it does offer is a few ways to find the line. The clearest one is this: the problem is less whether AI touched the application and more what the application is trying to do to the person, or the machine, reading it. At one end sits an obvious case. Greenhouse's survey found that 41% of job seekers use prompt injections, which are hidden instructions planted in a document to manipulate an AI screener Are job applicants and employers locked in an escalating AI arms race?. That isn't polish. It attacks the evaluator. It works for the same reason LLM judges reward fake references and fancy formatting whatever the content says: the reviewer's biases can be exploited without any access to its internals Can LLM judges be tricked without accessing their internals?.

A second line is about claims, not tools. Research on AI-written text about personal experiences finds it is false by its very nature. The model never had the experience, so a first-person story it invents has no real experience behind it, whatever anyone intended How does AI-generated false experience differ linguistically from human deception?. Applied to hiring, an AI-tightened description of a project you actually ran is editing. An AI-invented 'time I led a turnaround' is fabrication, even if no one meant to lie. Credentials have a related weak spot: recruiters gave candidates 8–15 points more interview invitations for listing AI skills, and certificates added little over simply claiming them Do AI skills help candidates get more job interviews?. When a claim isn't checked, the honesty of the claim is the only safeguard.

The less obvious part is that some deception may not feel like deception to the applicant. Studies of competence misattribution find that when AI output is smooth and the line between human and AI work blurs, people start believing they have skills they don't Do AI-assisted outputs fool users about their own skills?. Four mechanisms make this worse together: unclear credit, the illusion that fluent output means competence, handing off the thinking, and not seeing how the work was produced How do AI tools trick users into overestimating their own skills?. A candidate whose AI-polished portfolio overstates their ability may sincerely believe it. A useful test is whether the person who shows up to the interview, and then the job, can do what the application says. Separately, experiments show that people inclined to cheat prefer reporting to machines over humans, because lying to a form feels less costly Do dishonest people prefer talking to machines?. That suggests fully automated screening may itself invite more gaming.

The line cuts the other way too. Accusations of AI use often rest on nothing that actually separates AI text from human writing. They act as gatekeeping, and the harm lands on honest human writers Do unfounded AI accusations harm human writers instead?. Recruiters who say they can 'spot' AI deception (91% in the Greenhouse data) may also be flagging well-written human applications. The corpus suggests a partial way out: when people disclosed AI involvement, others were wary at first, but the wariness reversed once they saw consistent results Does revealing AI identity help or hurt user trust?. A workable rule might be to disclose the help, never manipulate the screener, and claim only what you can back up. The corpus doesn't test that rule in hiring, so treat it as a lead worth following, not a settled answer.