AI Trained to Spot Fake Job Ads That Lead to Forced Labour

Fake job ads are one of the main doorways into forced labour. A recruiter promises hotel work abroad, a visa, free housing — then takes your passport and your wages. Charities and job boards know this happens constantly, but nobody has had a reliable, fast way to sort the genuine listings from the traps. Scammers are also careful: their ads copy the look and feel of real ones, so they don't raise suspicion at a glance.

A research team set out to change that. They collected 464 verified job ads — 164 confirmed deceptive, 300 legitimate — through anti-slavery charities, covering nine countries of origin and 21 industries. Then they fed those ads to AI models that look at three things at once: the wording, the images, and the basic structure of the listing. This is where the technique matters. Most screening today relies on a human reading carefully, which is slow and inconsistent. The AI compares thousands of small signals in seconds.

The strongest giveaway wasn't anything dramatic. It was writing that wasn't quite professional: odd readability, clusters of risk-related words, and ads that leaned on visa sponsorship as a hook. Images mattered too — colours and textures that looked cheap or reused. The researchers' theory is simple: real employers have reputations and budgets to protect, so they invest in polished communication. Exploiters can't afford to be convincing across every channel at once, and that gap leaks through.

The team turned the findings into a trial tool that gives caseworkers a clear risk score, with reasons they can read and question. That last part is deliberate — the goal is to help a human decide faster, not replace them.

The honest catch: 464 cases is a small sample, the tool is a prototype, and the researchers admit it isn't deployed in the real world yet. Fraudsters will also adapt once they know what's being measured.