On September 5, 2026, The Guardian reported that food delivery riders are demanding platforms lift the lid on AI systems they say have depressed their earnings, turning opaque dispatch and pricing rules into everyday algorithmic pay cuts (The Guardian). The push now has academic backing, as researchers help riders interrogate the models that decide who gets which order, when, and for how much.
Why riders say algorithmic pay cuts are getting worse
Riders describe a familiar pattern. Base fees fall, per-mile rates shrink, and extra pay for bad weather or peak hours becomes rarer or conditional. At the same time, the apps introduce order batching and new metrics that quietly punish rejections or short breaks. Each tweak, on its own, looks minor. In aggregate, it feels like a pay squeeze the worker can’t see.
Those claims line up with years of data-rights work from groups such as Worker Info Exchange, which has documented how dispatch rules and driver scoring can move earnings without a clear paper trail. Their 2021 report chronicled how riders used data access requests to spot hidden penalties and shifts in incentive schemes (Worker Info Exchange). For workers, the core complaint isn’t just the outcome. It’s the opacity: you can’t contest what you can’t see.
How delivery app algorithms decide who gets paid
Platforms disclose little, but rider screenshots, court filings, and academic studies sketch a common playbook. Systems estimate delivery time, match jobs to nearby couriers, and weigh distance, traffic, and restaurant readiness. They also track acceptance rates, on-time records, cancellations, and time spent available. Small changes in those weights can cascade into algorithmic pay cuts even when headline fees appear unchanged.
Dynamic pricing adds another layer. An area might surge when demand outpaces supply, but surges can be throttled by pre-emptive incentives that pull in riders, then fade as capacity rises. Batching increases efficiency for the platform, yet it can turn into longer trips that pay less per stop. Without access to the logic, workers can’t tell whether a dip reflects demand, a personal score, or a model retraining that shifted payouts.
What the EU AI Act will require of gig platforms
That opacity collides with new law. The European Union’s AI Act classifies AI used for worker management and decisions affecting employment as high-risk, bringing strict duties on transparency, documentation, risk controls, and human oversight. Providers must document the model’s design and training data governance, ensure logs and traceability, and support meaningful information for affected people. Deployers must keep humans in the loop and assess risks to health, safety, and fundamental rights (European Parliament).
Timelines matter here. Core bans and governance provisions enter in stages, with high-risk obligations beginning to bite from 2026 onward across the bloc. That gives delivery platforms months, not years, to map where their dispatch, ranking, and pricing engines fall inside the AI Act, and to build the evidence trail regulators will expect. It also gives worker groups a calendar: a date when opacity stops being a product choice and starts being a compliance risk.
What riders and regulators can do now
Europe already gives workers leverage through data protection law. Under GDPR, riders can request copies of their personal data and seek “meaningful information about the logic involved” in automated decisions that produce legal or similarly significant effects. The UK’s Information Commissioner’s Office stresses that employers using AI must be clear about monitoring, ensure decisions are explainable, and provide routes to challenge outcomes (ICO guidance).
In practice, that means riders should ask for specific datasets and explanations: historic job offers and payouts, the factors that influence dispatch priority, any personal scores, thresholds for batching or surges, and how rejections or breaks alter future offers. Academics can help standardize templates and audit methods. Worker advocates have already shown that coordinated data access requests can reveal patterns, from deactivation triggers to pay “efficiency” tweaks that look like algorithmic pay cuts when plotted over time (Worker Info Exchange).
Regulators can move in parallel. They can require impact assessments for AI systems used in dispatch and pay, test whether explanations are specific rather than boilerplate, and sanction dark patterns that nudge workers into longer hours for diminishing returns. Under the EU AI Act’s high-risk regime, they will also be able to check technical documentation and logs, not just rider-facing statements. That’s where claims about fairness meet code.
What platforms risk if they ignore transparency
Compliance costs less than a scandal. A platform that can show how it prices jobs, how it monitors error rates, and when a human can override an assignment will be better placed with both regulators and workers. One that can’t risks fines under the AI Act, data protection penalties, and loss of supply in markets where riders can choose competing apps.
There’s also a business case. Clear rules reduce rumor-driven churn, help riders plan shifts, and cut support tickets that arise from guesswork. Companies already track these metrics. Sharing the logic behind them—safely and without exposing trade secrets—will be the test. Vague dashboards won’t cut it when enforcement lands and worker groups bring structured evidence.
The Guardian’s report shows the pressure building. The legal clock in Europe is ticking. Platforms that keep workers in the dark about dispatch, pricing, and penalties will face more than bad headlines. They will face audits, fines, and rider exits fueled by documented algorithmic pay cuts. The fix is straightforward: explain the system, measure its harms, and give people a human path to contest decisions—before regulators do it for them. For more on this, see reuters.com and bloomberg.com and nytimes.com.
Related reading: AI Update • Automation • Generative AI