Are job applicants and employers locked in an escalating AI arms race?

Chait states the loop in one passage: "Jobseekers use AI to apply to more and more jobs, while employers use it to filter candidates back out again. It's an AI doom loop that's getting worse, not better." The excerpt supports each leg with counts from Greenhouse's survey. On the applicant side, 49% of the 1,200 U.S. job seekers surveyed submitted more applications than a year ago, 69% have encountered fake job postings, 54% have encountered an AI-led interview, and 41% admit to using prompt injections, "hidden text designed to bypass AI filters." Of those who do not use the tactic, 52% say they are considering it. On the employer side, 91% of recruiters have spotted candidate deception, and 65% of hiring managers have caught applicants using AI deceptively: 32% reading from AI-generated scripts, 22% hiding prompt injections in resumes, and 18% appearing as deepfakes. The excerpt adds that 34% of recruiters spend up to half their week filtering spam and junk applications.

The mechanism the excerpt gives is escalation: each side's tool raises the other side's cost of acting. Candidates turn to hidden text because they see AI screening as opaque and unbeatable. Employers add filters because they are "drowning in so many applications" and are "looking for ways to sort through what's real and what's not." The excerpt's only evidence of direction is two one-year comparisons: 49% of job seekers submitting more applications than a year ago, and 74% of hiring managers more concerned about fake credentials, deepfakes, or misrepresented experience than a year ago. The excerpt contains no series over time, so "getting worse" is Chait's reading of those comparisons, not a measured trend.

This is the adversarial end of the overreliance problem in the Does AI augmentation protect workers from skill erosion? note. When both sides hand judgment to automated systems, oversight erodes in both directions, and the screening system stops being a reliable check on the applicant. It is also the case the Do university AI policies actually protect what credentials mean? note makes about credentials. A résumé or credential is a claim about a person, and once AI can produce the claim, a rule about which tools are permitted does not tell a reviewer whether the claim is true. Greenhouse's proposed remedies, identity verification and what the excerpt calls "good friction," try to restore that check outside the document. The excerpt reports that 36% of U.S. job seekers have used AI to alter their appearance, voice, or background in video interviews, which is the same problem seen from the candidate's side.

The excerpt does not model the loop or test its direction. It does not show whether the two legs are causally linked, whether employer filtering drives applicant gaming or the reverse, or whether the trend has persisted beyond the one-year comparisons. "Deception," "deceptively," and "fake" are not defined, and the 41% figure is an admission in a survey, not a count of actual use. The survey's method is also unstated: no field dates or sample sizes for the recruiter and manager figures. The loop is best read as a hypothesis that the excerpt's numbers are consistent with, not a demonstrated feedback process. Testing it would take the same measures repeated over time, with both sides' behavior recorded in comparable units.