Amazon used an artificial intelligence tool called Atlas to assess the likelihood of workers unionizing across more than 50 warehouses in Canada, the United States and Europe, according to internal company documents obtained by The Globe and Mail and published on October 23. The records surfaced during hearings between Amazon and the union Unifor at the British Columbia Labour Relations Board. The tool analyzed hundreds of data points to build what Amazon described as a labour risk framework, rating facilities on a 1-to-5 tier scale and a 100-to-900 score, where lower scores signaled higher unionization risk.
How Atlas scored the warehouses
Atlas rated facilities on two scales. The primary scale ran from 1 to 5: a score of 1 signified reported workplace disruptions, while 5 meant no disruption risk. The secondary scoring system ranged from 100 to 900 points, where higher numbers indicated a lower likelihood of a union vote. The documents tracked high-severity employee relations events, including card signings, protests and work stoppages.
The YVR2 warehouse in Delta, British Columbia, which unionized under Unifor in July 2025, received a Tier 1 ranking with a labour risk score of 169. YVR2 recorded 52 high-severity events, while the nearby YVR3 warehouse registered 47. According to earlier reporting by openDemocracy on the same documents, two Amazon warehouses in the UK and 14 in the US and Canada were placed in the highest-risk tier.
What the system measured
The inputs fed into Atlas covered safety incidents, leadership availability and training completion, according to Amazon spokesperson Sam Stephenson, who said the system gathers that information “to spot trends”. Stephenson added that human teams make all final decisions rather than automated systems. He did not say whether Amazon currently uses Atlas or similar tools in its Canadian operations.
The disclosure has already drawn fire. Britain’s GMB union called the tool “Orwellian” and said the public deserves to know whether Amazon uses Atlas-style systems in the UK, where two warehouses were placed in the highest-risk tier. Earlier reporting on the same documents described Atlas ingesting hundreds of data points, including worker feedback channels, training records, health and safety incidents, tenure and turnover, and local economic conditions. The framework was presented internally as a way to evaluate “emerging workforce issues, labour union targets and internal or external labour threats”.
Why It Matters
The Atlas documents land in the middle of a live legal fight. The records emerged from Amazon’s hearings with Unifor at the British Columbia Labour Relations Board, and the scoring of a warehouse that had already unionized under that same union gives the story its sharpest edge. Regulators and courts are increasingly asking whether workforce-analytics tools cross the line from management insight into surveillance, and a 1-to-900 scoring system for union risk is exactly the kind of system that invites that question.
The story also widens the frame on how AI is used inside big tech. While Amazon’s wider AI stack, from Alexa+ to Bedrock, is marketed as customer and developer products, Atlas shows the company’s internal AI pointed at its own workforce. As state attorneys general are already suing AI companies over safety claims, the regulatory pressure on how AI systems are deployed is only growing.
Background
Amazon has faced years of unionization efforts at its warehouses, including the YVR2 facility in Delta, British Columbia, which joined Unifor in July 2025 — the same warehouse Atlas had scored Tier 1 with a 169. The company invests heavily in AI across its operations, from the Alexa+ assistant to the Bedrock model platform. The Atlas documents are the clearest look yet at how that AI investment extends to labour relations.
Source: The Globe and Mail, via el-fondo, openDemocracy investigation, via Portside
0 Comments