Amazon Uses AI to Score Union Risk at Its Warehouses, but Denies Using It in the UK

Amazon reportedly uses AI to assign each of its sites a score for how likely workers there are to unionize or take industrial action. On October 16, the British outlet openDemocracy described how a system called Atlas works, based on internal Amazon training documents. The documents surfaced during a legal dispute between Amazon and the Canadian union Unifor. They include risk ratings for UK sites, though Amazon says it does not currently use Atlas in Britain.

How employee surveys become a score

According to the report, Atlas pulls together several kinds of information to assign each site a labor activity risk score. Hundreds of inputs are said to feed into it, including employee survey responses, Reddit posts, conditions at the site, and any history of past disputes. In other words, the system factors in not only the feedback workers give the company directly but also online posts and the site’s past record.

The lower the score, the higher the assessed likelihood of unionization or industrial action. Scores range from 100 to 900, and sites are sorted into five tiers. Amazon has described Atlas as a system for understanding, at the site level, the concerns employees are raising.

The outlet reported that 16 sites in the United States, Canada, and the UK appeared in the highest risk tier. In the UK, those were BHX4 in Coventry and EMA4 in Sutton Coldfield. Amazon has said only that it does not currently use the system in Britain; it remains unclear when and how it may have been applied there in the past.

Spotting problems, or monitoring unions?

Amazon and the union give conflicting accounts of what Atlas does. Amazon says it is not a tool for tracking individual employees’ views, their union membership, or their personal data. The union, by contrast, says Amazon uses Atlas to monitor union organizing.

BHX4 in Coventry, which appears in the documents, is a site where a union recognition process actually took place. According to a ruling by the UK’s Central Arbitration Committee, the union GMB pushed for a recognition ballot there in 2024 but fell short of the threshold required for recognition.

AI at work: what to watch

For working people, the most immediate question raised by this controversy is what their own answers and workplace information end up being used to decide. The report that employee surveys feed into labor activity risk ratings, in particular, makes the purpose of the information the company collects a point of contention. That is where Amazon’s account—that it is identifying employee concerns—meets the union’s claim that it is monitoring union activity.

For Korean readers, the case is also a window into how AI handles employee-related information in the workplace. What to watch next is when the UK site ratings were produced, how they were actually used, and how accurate the risk predictions turned out to be. Whether such scores genuinely influenced site operations or decisions affecting workers is another open question.