Garfield Hylton, an Amazon warehouse worker in the UK, has long known that the company records every minute of his working day.
What he probably did not know is that his warehouse could also be given a score.
What the documents reveal
Internal Amazon documents were made public during a dispute with a union in Canada, in which the local labour relations authority ruled that the company had broken several laws. They appear to be used to train managers.
They describe an AI tool called Atlas. Its role: to estimate, for each warehouse, the likelihood that employees will organise, by forming a union or taking collective action.
To do this, Atlas is said to gather hundreds of pieces of information: employees' answers to internal surveys, messages on the company's internal forum, Reddit posts, whose recurring words and the reactions they provoke it analyses, but also workplace safety data, pay levels, or proximity to other sites that are already unionised.
A score from 100 to 900
Each warehouse receives a score between 100 and 900. The lower it is, the more Atlas estimates that employees are at risk of organising. Sites are then sorted into five categories, from "no immediate risk perceived" to "imminent action".
According to the documents, 16 warehouses in the United States, Canada and the UK fall into the highest category, including two British sites near Coventry and Birmingham, where employees are in fact running a campaign to win union recognition.
The reaction
The British union GMB condemned a practice worthy of George Orwell, and called on the government to demand answers. A worker at the Coventry warehouse restated the obvious: it is for workers to decide whether they want to unionise.
Amazon says a system called Atlas is not currently used in the UK, and that the tool operates at site level, not at individual level. The company has always maintained that it does not want to prevent its employees from joining a union. Legal experts interviewed believe these documents cast doubt on that claim.
In France, such a tool would run into two serious obstacles. Union membership is among the data that the European data protection regulation classes as most sensitive, and processing it is in principle prohibited. And the Labour Code prohibits an employer from taking union activity into account in its decisions, whether on hiring, promotion, discipline or work organisation. Freedom of association is also a right enshrined in the preamble to the Constitution.
What it says about AI at work
We often hear about AI monitoring productivity. Atlas goes further: it does not measure what employees do, it tries to predict what they might do together.
It is a new imbalance. To organise, employees need to talk to each other, often discreetly. An AI that reads the weak signals of those conversations gives the employer a head start. It is not the first case of AI turning workers into data, as shown by these Indian workers who film their movements.
Key takeaways
An AI capable of anticipating a collective movement is not neutral. It gives whoever owns it a new power over those it observes.
The question is not only whether Atlas is legal here or there. It is what kind of workplace we want, when a machine can guess what employees are saying to each other before they have even decided it.