Court-disclosed documents describe an AI-assisted framework called Atlas that assigns covered Amazon warehouses a score between 100 and 900, with lower numbers flagging a higher perceived risk of union organizing or labor disruption. Two UK facilities reportedly landed in Tier 1, the classification described as reserved for sites where worker action is considered imminent.
The documents surfaced through litigation between Amazon and Canadian union Unifor. They describe a five-tier system built on hundreds of data points, according to openDemocracy and Novara Media reporting. Reported inputs include employee feedback, safety incidents, turnover rates, previous strike action, local economic conditions, and activity on external platforms including Reddit.
What Atlas Reportedly Does, and What Amazon Says It Does Not
The gap between Amazon’s “employee support” framing and the system’s apparent risk-scoring logic is where the real dispute lives.
BHX4, near Coventry, reportedly carries an active organizing campaign, while EMA4, in the Sutton Coldfield or Minworth area, has seen previous strike action. Both are identified in reporting as Tier 1 sites. Fourteen additional high-risk sites are reported across the United States and Canada.
Amazon denies that Atlas tracks individual workers’ opinions, affiliations, or union membership. A spokesperson described the system as operating at the site or building level, intended to identify where employees need support and ensure workplace concerns are addressed.
Amazon reportedly told one outlet the framework is no longer called Atlas and is not currently deployed in the UK. That claim directly conflicts with the interpretation that the disclosed documents show active UK use, so the UK deployment question remains contested, not confirmed.
A site-level risk score does not need to name individual workers to shape decisions that affect every single one of them. A system that ingests worker feedback, public Reddit posts, and labor-action history to produce a facility’s “risk” classification can still influence staffing, resource allocation, and management responses across an entire workforce.
The Accountability Gap Nobody Has Closed
The publicly reported materials leave critical questions unanswered, and workers have no apparent mechanism to challenge how their workplace was classified.
The publicly reported materials do not establish how accurate Atlas’s predictions are, how long the data is retained, who audits the model, or how a workforce could challenge a site’s classification. Those omissions matter. They echo broader concerns raised by cases of secretly tracking users without their knowledge or any meaningful avenue for recourse.
Michael Newman, a partner at law firm Leigh Day, observed in Novara Media reporting that the scoring system’s own framing is telling: higher scores are described as better for company leaders. That framing does not suggest a neutral employee-support tool. It suggests a system calibrated to treat organizing as an operational liability.
The GMB union characterized the alleged tools as “Orwellian” and demanded public clarification about whether Atlas or an equivalent system is currently running in UK warehouses. The documents also reportedly instruct managers to listen to workers’ concerns without “over-sympathising,” a phrase that reads less like a wellness policy and more like crisis-communication coaching.
If you work in a warehouse, the question is not whether Amazon has a file with your name on it. The more pressing question is whether the facility around you has been scored, tiered, and earmarked for pre-emptive management intervention before any collective action takes shape.
Independent auditing of systems like Atlas is not a radical demand. Regulators have already moved to restrict Amazon and Google from handling sensitive data without sufficient oversight. Neither is meaningful worker access to challenge site-level classifications, nor transparent disclosure of what data feeds a labor-risk score. Until those mechanisms exist, the distance between “employee support” and “union suppression” remains exactly as wide as Amazon’s current position requires.
