The Hiring Process Is Now a Bot Fighting Another Bot and Real Candidates Are Losing in the Middle

The arms race that nobody announced

Nobody held a press conference to announce that hiring was becoming a bot-versus-bot contest. It happened through a series of individually rational decisions that collectively produced an irrational outcome.

Recruiters started using AI to screen resumes because application volumes were becoming unmanageable. Candidates started using AI to write resumes because AI-screened resumes rewarded specific language patterns that human writers struggled to replicate consistently. Recruiters added AI detection tools because the flood of optimised resumes made it impossible to tell which candidates had the skills they claimed. Candidates added prompt injections to beat the detection. Recruiters added AI-assisted interviews to verify skills. Candidates used AI to prepare for AI interviews. Each escalation was rational for the individual actor and catastrophic for the system as a whole.

The outcome is documented. LinkedIn processes 11,000 applications per minute, a 45% increase from the year before. 67% of HR leaders say AI-generated applications have slowed their hiring process. 53% of job seekers were ghosted by an employer in the past year, a three-year peak. Time-to-hire is up. Trust is down. 34% of recruiters now spend up to half their working week filtering spam that the AI was supposed to eliminate. The efficiency gains AI promised to deliver to hiring have been entirely consumed by the overhead of managing the arms race it created.

The doom loop has two layers now

The conventional story about AI resumes was that they helped candidates get past the initial ATS screening by matching keywords the algorithm rewarded. That story was already creating problems. The new story is more specific and more brutal.

A resume now has to pass two filters that point in opposite directions. The first filter is the AI screener, which prefers structured, keyword-dense, well-formatted resumes that AI systems are good at producing. The second filter is the human hiring manager, who has started categorically rejecting resumes that read like they were written by AI. Internal estimates at large enterprises suggest 60 to 80 percent of resumes received in 2026 show clear signs of LLM authorship. The phrase this reads like ChatGPT has become a one-word rejection note in hiring committees.

Both filters have to pass and they favour opposite things. A resume optimised to clear the AI screening is more likely to be rejected by the human. A resume written to sound authentically human may score lower on the AI screening. The candidates getting through are not the ones who avoided AI or the ones who used it most aggressively. They are the ones who understood both layers well enough to navigate between them, a skill that has nothing to do with whether they can actually do the job.

What 91% of recruiters suspecting deception actually means

Greenhouse's 2026 AI Hiring Report found that 91% of recruiters and hiring managers have spotted or suspected candidate deception. The specific deceptions they describe go well beyond AI-polished resumes. 48% report seeing fake references. 35% have encountered candidates using AI during live interviews. 31% have encountered candidates in different time zones than stated on applications. 31% have had a different person show up for interview than the one who applied. 18% have encountered deepfake video interviews where the person on screen was not the person being hired.

These are not edge cases from a paranoid minority of recruiters. They are the majority experience. The hiring system is now operating in an environment where the default assumption is that something in the application may be fabricated. That assumption is corrosive to every genuine candidate in the pipeline, because verification overhead now applies to everyone regardless of whether they have done anything deceptive.

The candidate-side response is its own data point. 46% of job seekers say their trust in the hiring process has decreased in the past year. 42% attribute that decline specifically to AI use. 87% want employers to be transparent about how AI is used in hiring. The mutual suspicion is not one-directional. Both sides are experiencing the same system as untrustworthy for different reasons, and neither side has the individual leverage to fix it unilaterally.

The candidates losing cleanest

The most revealing version of the doom loop involves the candidates who did everything right and lost anyway.

A candidate who wrote their resume honestly, without AI, gets scored lower by an AI screening system trained on the optimised keyword density of AI-polished resumes. They lose to candidates who used AI more aggressively despite having equivalent or superior actual qualifications. The tool that was supposed to surface talent more efficiently is sorting away from the thing it was designed to surface.

A candidate who used AI to better express genuine experience gets flagged as an AI writer and rejected before reaching a human reviewer who might have recognised the underlying qualification. The tool that was supposed to catch deception is catching honesty that was expressed in a particular way.

Stanford HAI research documented that AI resume detectors produce false positive rates exceeding 20% on non-native English writers and creative writing. A candidate who writes in English as a second language or who uses an unconventional writing register is statistically more likely to be flagged as an AI writer than a native speaker writing in a standard register. The detection tool that was supposed to ensure fairness is introducing its own form of bias that disproportionately penalises specific groups of legitimate candidates.

Where this is heading

The structural fix that hiring researchers most consistently point toward is skills-based hiring: replacing the resume as the primary evaluation artifact with direct demonstrations of the specific skills a role requires. Portfolio reviews, work samples, standardised skills assessments, and structured references all move evaluation toward what a candidate can actually do rather than what an AI can produce a polished description of.

Skills-based hiring produces measurably better outcomes when implemented well. Companies using it report better first-year retention, stronger role-fit, and reduced time spent on candidates who interview well but cannot do the job. It is also genuinely difficult to implement at scale because it requires rewriting job descriptions, investing in evaluation infrastructure, and changing the cultural assumptions about what credentials and experience signals mean.

The pace of adoption of skills-based hiring and the pace of the arms race are not matched. The arms race is moving faster. The hiring system is getting worse at a rate that the structural fix cannot currently keep up with, and neither side has an obvious individual exit from the loop they are both stuck inside.