AI Hiring Tools Reward Who You Know — Researchers Say That's Unfair

A team of eight researchers from universities in Europe, the US and Chile published a "perspective" paper — an argument backed by existing evidence, not a new experiment. Their claim: the AI tools used to screen job applicants, suggest connections or rank candidates are trained on social networks, and those networks already favour people with the right friends, mentors and colleagues. When software learns from that data, it can lock the advantage in.

Their example is academic hiring. Two researchers can have identical publication records, but one comes from a famous lab with deep connections. Hiring committees read that as "better fit", and any AI trained on past hires learns the same lesson. The authors identify ten network effects — including the tendency for similar people to cluster together, and for friends-of-friends to become friends — that quietly distort what employers think they are measuring.

The paper's bigger point is about how we check for fairness. Most audits compare outcomes between groups: what share of women or minority candidates got hired. That is useful but incomplete, the authors argue, because it ignores how the decisions were actually made. A fair-looking result can still come from a biased process. They call for judging both the outcome and the procedure, and for involving the people affected in deciding what counts as fair.

The catch: this is a call to action, not a tool. The paper offers no software, no fairness score and no easy checklist for companies. It also notes that networks are not automatically bad — connections help everyone, and strong networks can lift people up. The real question is whether a shortcut is legitimate. For now, employers and AI vendors have a way of thinking to work with, not a fix to install.