180 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 30 August 2026
← The incident library
HD-INC-113
Food delivery · Italy · 2021 · Discriminatory ranking algorithm

A court in Bologna found that Deliveroo's rider-ranking algorithm discriminated against couriers because it treated a legally protected strike the same as an unexplained no-show

By Ellie Harris · Filed Booking model in use to 2020; ruling reported January 2021

Alleged: Deliveroo developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

A court in Bologna found that Deliveroo's rider-ranking algorithm discriminated against couriers because it treated a legally protected strike the same as an unexplained no-show

What happened

It was reported that Deliveroo’s booking system in Italy relied on a reputational algorithm, referred to internally as Frank, that scored couriers using statistics about how they worked, including how reliably they turned up for the sessions they had booked, and that riders with better scores were given earlier access to choose the shifts they wanted. The case, brought by three unions affiliated with the CGIL confederation, turned on a specific feature of that scoring: the system treated a rider who missed a booked session because they were ill, or because they were taking part in a strike, the same as a rider who simply failed to appear, so that an absence for a legally protected reason lowered the score in exactly the way an unexplained one did.

In a ruling reported in early January 2021 the Court of Bologna found this discriminatory. A rule can look even-handed and still hit one group harder than another, and this one, by counting a strike or a sick day as just another missed shift, punished the riders who took part in a strike or who could not work because they were ill. The court ordered Deliveroo to pay 50,000 euros to the unions that brought the case, cover their legal costs, and publish the ruling on its website. The CGIL called it a turning point for workers’ rights in the digital economy. Deliveroo said the ranking belonged to an old booking system it no longer used in Italy or anywhere else, and that riders were free to choose when they worked.

What an auditable version would have shown

The court could see the rule was unfair just from how it was built. What no one could say is how many riders it actually cost, because Deliveroo kept no record tying each missed shift to the reason behind it. A system built to be checked would log, for every booking, why a rider missed it and how far their score fell as a result, so the number of people marked down for striking or being ill, and the size of the hit each time, would be a figure you could pull up rather than a point fought over in court.

Where the gap was

Missing a shift dropped a rider’s score, the score decided who got first pick of the good shifts, and why the shift was missed, a strike, a sick day, or simply not turning up, made no difference to how it counted. Two records would have changed that. A MetricRecord would track how the ranking treats riders according to why they miss work, so a pattern of punishing strikes shows up as a number inside the company rather than something a union has to prove in court. A ConductRecord would keep each score change and the reason behind it, so a mark against someone for a protected absence can be found and undone. Neither of these makes the underlying choice, to treat a strike like a no-show, which is Deliveroo’s to make or not. They make it visible, and reversible when it goes wrong.

What governance should have looked like

When a score decides how much work a person gets, it has to tell a protected reason for missing a shift apart from an ordinary one, its effect on things like strikes has to be something you can measure, and a rider has to be able to see a mark against them and challenge it. The court reached its finding from the way the system was built, without having to count the damage. A proper record is what lets you count it, and give back what it took from each rider it wrongly marked down.

Failure Pattern: a reputational ranking scored workers on reliability without separating legally protected absences from ordinary ones, so exercising a protected right lowered a score that governed access to work, and nothing measured that effect by the reason for the absence.

Governance Principle: where an automated score governs a person’s access to work, it must distinguish legally protected reasons for absence from ordinary ones, and the effect of the score on protected activity must be measurable and recorded.

The reference implementation of MetricRecord and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Deliveroo could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →