The use of artificial intelligence in recruitment is growing rapidly, with tools being deployed to automate everything from resume screening to interview scheduling. But as we look at how AI detectors are being used in education to flag potential cases of academic dishonesty, there are important lessons to be learned—especially when it comes to the risks of deploying similar technology in the recruitment process.
A recent Bloomberg article highlighted how AI detectors in schools are creating false accusations, particularly affecting neurodivergent and ESL (English as a Second Language) students. As AI implementors, we must consider the implications of these developments for recruitment.
While the promise of AI to streamline the hiring process is enticing, the risks of using text classifiers or AI detectors in screening candidates could lead to unintended bias, false accusations, and serious consequences for both candidates and employers.
The Problem of False Positives: When AI Gets It Wrong
In the academic world, AI detectors are used to determine whether a student’s work is likely to have been written by a generative AI tool. However, as the Bloomberg piece shows, these detectors are far from perfect. False positives—where work written by a human is wrongly flagged as AI-generated—are common. One case cited involved a student with autism whose formulaic writing style was flagged by AI detection tools, jeopardising her academic standing despite the work being her own.
Imagine this same scenario applied to recruitment. An AI detector flags a candidate’s resume or cover letter as being AI-generated simply because their writing style doesn’t align with the algorithm’s expectations. The result? A potentially qualified candidate is excluded from the hiring process for something they didn’t do. In recruitment, the stakes are incredibly high—career opportunities, livelihoods, and the future of organisations depend on getting the right hire. False positives could easily become false eliminations.
Bias Against Non-Traditional Candidates
One of the most concerning aspects of using AI detectors in recruitment is the risk of reinforcing existing biases. As with students who write in a more generic or structured way, candidates who speak English as a second language or who come from neurodiverse backgrounds are more likely to be flagged by AI tools. This echoes findings from the education sector, where students with certain backgrounds are disproportionately affected by AI detectors.
Recruitment is already fraught with concerns over unconscious bias, and AI detectors could exacerbate this problem by unfairly disadvantaging candidates who don’t fit a narrow mould. Candidates from diverse cultural and linguistic backgrounds, or those with different cognitive processing styles, could find themselves penalised not because they lack the necessary skills but because their writing style triggers AI suspicion.
The Arms Race of AI Detectors and AI Circumvention Tools
In response to AI detectors, a new wave of tools is emerging that help people “humanise” AI-generated content to pass through detection systems. In the recruitment world, this could lead to an arms race between AI detectors and candidates using AI to evade detection. Candidates may start using tools to bypass AI screening, rewriting their AI-generated cover letters or CVs in ways that are undetectable by classifiers.
In fact, for somebody who can prompt reasonably well, a simple addition to your prompt of "write with a high degree of perplexity and burstiness" will usually circumvent any AI detector.
This is an unproductive cycle. Rather than improving the recruitment process, this escalation undermines the integrity of the system. It shifts the focus from evaluating a candidate’s actual skills and potential to outsmarting detection tools. Recruitment should be about finding the best person for the job, not who can best navigate a flawed system.
Why Relevance and Context Matter More Than AI Detection
The key takeaway from these examples in education is that while AI detectors might seem like a solution to the problem of AI-generated content, they come with significant risks. In recruitment, this means that the blanket use of AI detectors to screen applicants is fraught with danger. Instead, we should be focusing on relevant AI—AI that is used where it truly adds value, enhances fairness, and drives meaningful outcomes.
Recruiters need to be cautious when implementing AI detection tools. Rather than relying on these tools as the sole gatekeepers in the hiring process, it’s essential to ensure that human oversight and context remain central. A candidate’s writing style, format, or even structure should not be grounds for exclusion, particularly when it doesn’t correlate with their ability to do the job.
Conclusion: The Need for a Human-First Approach
As AI becomes increasingly embedded in recruitment, we must learn from the missteps seen in education. Using text classifiers or AI detectors as blunt instruments to screen candidates risks excluding talented individuals based on flawed algorithms. In the rush to automate, recruiters must remain vigilant and ensure that their tools are both relevant and fair.
The stakes are simply too high to leave these decisions entirely to machines. Just as in education, recruitment needs a balanced, human-first approach where AI complements rather than dominates the process. Let’s ensure that we’re not just using AI for the sake of it, but in a way that truly benefits both organisations and candidates alike.