Amazon faces scrutiny over Atlas, an artificial intelligence system tool described in an internal document as assessing the risk of union activity at its warehouses. The document, disclosed during legal proceedings involving Canadian union Unifor, lists two UK sites in its highest-risk category.
The document appears to be training material for Amazon managers and describes a framework for evaluating emerging workforce issues, union activity and what it calls internal or external labour threats.
The GMB has called for transparency over whether Atlas or similar systems are used in the UK. Amazon denies that Atlas is intended to counter union activity and says it assesses activity at site level rather than monitor individual employees' union membership or opinions.
Amazon's Atlas AI Assigns Warehouses Union Risk Scores
The document describes Atlas as a 'labour risk framework to clarify how Amazon evaluates emerging workforce issues, labour union targets and internal or external labour threats.'
According to the document, Atlas assigns warehouses scores ranging from 100 to 900 to assess the risk of union activity. Lower scores indicate a greater predicted likelihood of workers organising, while higher scores are described as 'better' for company leaders.
Amazon has been accused of using AI to track employees in the UK and elsewhere and rank how likely they are to form unions.
— Novara Media (@novaramedia) October 19, 2026
The online retail giant uses a tool called Atlas in an apparent bid to predict acts of dissent by workers in the UK, the US and Canada, a document released… pic.twitter.com/VFbt9g741Q
The document also groups sites into five risk tiers. Tier 1 is associated with the assessment that 'future activity is imminent,' while Tier 5 indicates 'no immediate perceived risk.' These are the document's classifications, not independently verified predictions of what workers will do.
The document says the risk tiers can inform staffing decisions, resource planning and measures to mitigate risks to Amazon's labour operations.
The available reporting does not establish how accurately Atlas predicts union activity, how its scores are used in practice or whether they have led to disciplinary or other adverse action against workers.
Two UK Warehouses Listed in Atlas's Highest-Risk Tier
The document lists two UK warehouses in Tier 1, alongside 14 sites in the US and Canada, bringing the total in the highest-risk category to 16. The material does not establish whether those classifications remain current.
The document associates BHX4, near Coventry, with an 'active organising campaign' and lists EMA4 among the sites in Tier 1. The facility's precise location and the reported history of strike action should be confirmed against the original document before publication.
The listings do not independently establish that Atlas is currently operating at either British warehouse. Nor do they demonstrate that Amazon took action against workers because of union activity at those sites.
Amazon told openDemocracy that Atlas is not currently used in the UK.
Michael Newman, a partner at Leigh Day, said the findings presented 'a very different story' from Amazon's stated position on workers joining unions. The document does not, by itself, establish whether the scores were used to discourage union activity at either UK site.
GMB Seeks Answers Over Amazon's Use of AI to Assess Union Risk
The GMB has called for greater transparency over whether Atlas or similar systems are being used in the UK.
'Workers deserve to be able to work free of the fear of excessive surveillance,' said Rachel Fagan, a GMB organiser. 'If Amazon bosses are using Atlas or similar AI-fuelled systems in the UK, the British public deserve to know.'
The manual also advises managers to hear worker concerns without 'over-sympathizing,' though the guidance doesn't prove complaints were ignored or handled unfairly.
Amazon denies using Atlas to counter unionization, saying it assesses sites rather than individuals and helps identify where employees need support. The document outlines the system's intended uses but does not establish its accuracy, current UK deployment or any unlawful conduct.
