The hidden gatekeeper: AI in graduate hiring

Behind the scenes, automated tools are filtering CVs and scoring interviews, with real consequences for fairness.

Article link: https://open.substack.com/pub/johanosteyn/p/the-hidden-gatekeeper-ai-in-graduate?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Audio summary: https://youtu.be/VWBzf0WO2Z0

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

If you are a student or a recent graduate, you may think the toughest part of getting hired is competing with other candidates. Increasingly, the first competition is with a machine. Many employers now use automated systems to sift CVs, rank applicants, and even score video or online assessments. This is sold as efficiency, consistency, and “better matching”. But it also creates a new kind of gatekeeping: decisions that feel final, yet are hard to question, understand, or appeal. 

In a country like South Africa, where youth unemployment is already a national emergency, we cannot afford to add another invisible barrier to the system. The issue is not that technology is evil. The issue is that we are allowing high-stakes decisions to become less transparent, less accountable, and potentially less fair.

CONTEXT AND BACKGROUND

Recruitment has always been a filtering exercise, but the scale has changed. Large organisations receive thousands of applications for entry-level roles, and the temptation to automate is obvious. Today’s tools range from CV “parsers” that extract and standardise information, to assessments that infer skills, and systems that recommend who should be shortlisted. In practice, this means many graduates are never seen by a human being.

Globally, regulators and courts are starting to notice. A useful signal is how much attention is now paid to algorithmic bias in employment and the limits of existing enforcement tools, as discussed by the Associated Press in the context of shifting civil-rights enforcement approaches. In other words, society is waking up to the fact that automated hiring does not remove discrimination risk; it can simply hide it behind software.

INSIGHT AND ANALYSIS

Here is the uncomfortable truth: AI systems do not “understand” people. They detect patterns. If historical hiring data reflects bias, the model can learn that bias. If the system relies on proxies such as university names, locations, gaps in employment, or writing style, it may penalise precisely the candidates we most need to uplift. This is why legal scrutiny is increasing around whether vendors and employers can escape responsibility by saying “the tool didn’t make the final decision”. Reuters has reported on the growing role of state and local rules and litigation in the US as AI plays a bigger part in employment decisions.

Then there is the “voice problem”. If an assessment includes speech-to-text or video analysis, accent and dialect can become a hidden disadvantage. Axios recently highlighted how AI systems can struggle with non-standard accents and how that matters in high-stakes settings like hiring. In South Africa, where language and accent diversity are normal, this should set off alarms.

Finally, we must talk about the psychological impact on young people. An automated rejection can feel more definitive than a human one: you cannot phone the algorithm. You cannot ask what you did wrong. You cannot learn. That creates a corrosive mix of anxiety and cynicism, at exactly the point where we should be building confidence and capability in the next generation.

IMPLICATIONS

For employers, the first step is honesty. If you use automated tools, say so plainly. Explain what they evaluate, what they do not evaluate, and how humans will review results. The Financial Times recently captured how inconsistent organisational rules around AI create confusion and risk inside companies. The same logic applies to hiring: unclear systems create distrust and reputational damage.

For policymakers, we need a pragmatic approach: transparency requirements, documented bias testing, and meaningful routes for candidates to query outcomes. The Washington Post has covered how states are moving towards regulating AI in employment, driven partly by high-profile legal cases and growing concern about discrimination. South Africa should not copy-paste foreign law, but we can learn from the direction of travel.

For graduates and parents, the practical advice is not to “game the system” but to be aware of it. Build a clear, skills-based CV. Keep evidence of your work: projects, portfolios, references, and outcomes. Where possible, apply through channels that allow a human touchpoint. And when you are rejected repeatedly with no explanation, it may not be your capability that is being measured.

CLOSING TAKEAWAY

AI in hiring can be useful, but only if we treat it as decision support, not decision replacement. The real danger is not automation itself; it is the quiet erosion of accountability. When young people are filtered out by systems they cannot see, we create a workforce that feels powerless before it even begins.

South Africa’s future depends on widening opportunity, not narrowing it through invisible gates. The question we should be asking is simple: if an algorithm shapes a young person’s career prospects, who is responsible for ensuring it is fair, explainable, and open to challenge?

Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net