Europe has a rule: before a government agency or company uses high-risk AI to decide who gets a job or who gets public benefits, they have to check if it could trample your rights. That check is called a Fundamental Rights Impact Assessment. And according to a September 2026 preprint, the evidence you'd need to do it credibly is scattered across incompatible databases, legal texts, and risk vocabularies. It's like trying to assemble IKEA furniture when the instructions are in six languages and half the screws are missing.
The paper, by Faith Olopade and two co-authors, proposes a fix: a reusable Semantic Web framework that pulls that evidence into one searchable place. They demonstrate it on two areas where the stakes are highest — employment and worker management, and access to essential public services. Both target public bodies. Both have real victims: job applicants, benefits claimants. This isn't academic. It's about whether your next rejection letter comes from a human or a black box that never had to show its work.
But here's the twist. When the researchers tested AI to automatically sort that evidence, it performed near random. Let that sink in. The same technology that companies want to use to judge fairness can't even reliably find the evidence of past harm. It's like hiring a referee who can't see the foul. That's not a reason to abandon AI — it's a warning against letting machines judge fairness alone.
The timing is not subtle. Article 27 of the EU AI Act is now biting for public-sector deployers. A July 2026 survey in ScienceDirect lists FRIAs alongside compliance checks and incident reporting as tasks that Semantic Web technologies could potentially address. And a January 2025 paper — claiming the first ontology for FRIAs — shows researchers are racing to build the infrastructure. This is not a solo act. It's a land grab for who defines how Europe enforces AI rights.
Why should you care? Because if the evidence base is broken, the assessments are theater. Companies could tick boxes while algorithms quietly deny benefits or filter out job candidates. The paper is a v1 preprint — not peer-reviewed — so treat it as a research program, not a standard. But the warning is real: automation without solid evidence is just bias with a receipt.
The fix isn't sexy. It's ontology work, data integration, legal mapping. But it's the difference between rights on paper and rights in practice. Watch whether public bodies actually adopt these frameworks — or whether they keep using AI as a rubber stamp. If they do, the AI Act might become the most well-documented failure in tech history. Your next job application could depend on it.