Threat assessment: high. Amazon uses an in-house AI forecasting tool called Atlas to predict how likely workers at its warehouses in the US, Canada and the UK are to unionise, stage a sit-in or take other “disruptive labour events,” Hacker News reports. Internal documents came out during a court fight with Canadian union Unifor. They show Atlas scoring warehouses on union risk, and two UK sites sit in the top danger tier.
This is one of the clearest examples so far of a major employer pointing predictive AI at its own workforce’s right to organise.
🎯 Situation Report
The documents look like training material for Amazon managers in Canada. According to the report, an appendix says Atlas collects “hundreds of data points” to give each warehouse a score from 100 to 900.
Known inputs:
- Posts on Reddit forums
- Employee responses to internal feedback surveys
- Wider workplace data
- Any past industrial action at the site
A lower score means Atlas thinks organising is more likely. The document says that, for leadership, higher scores are “better.”
📊 The Tier System
Atlas groups warehouses into five tiers:
- Tier 5: “no immediate perceived risk”
- Tier 1: “future activity is imminent”
The document says the scale can “inform staffing decisions, resource planning, standard actions, and ownership to mitigate risks to Amazon labour structure.”
16 warehouses are rated Tier 1. Two of them are in the UK:
- EMA4: a £500m robotic fulfilment centre in Sutton Coldfield, north-east Birmingham
- BHX4: the warehouse near Coventry
BHX4 is no random pick. Its workers were the first in the UK to strike against Amazon, in January 2023. Six months later, around 2,600 of them voted in an official ballot for GMB union recognition. GMB lost by just 28 votes.
🛡️ Amazon’s Position
An Amazon spokesperson pushed back on calling Atlas “individual employee monitoring or surveillance,” saying: “This is site/building level activity only.” The company says it doesn’t track individual opinions, affiliations, union membership or “personal data.” It describes Atlas as a tool to “help us understand where associates need more support.”
The site-level defence will get tested. If a model reads Reddit posts and survey answers to flag one building as “imminent,” management attention lands on the people inside it, whether or not anyone is named.
⚔️ The Opposition
Michael Newman, a partner at law firm Leigh Day, said the scoring “makes it clear that they think the union is bad.” He added that it tells “a very different story” from Amazon’s earlier claims that it doesn’t try to stop workers from joining a union.
GMB organiser Rachel Fagan was blunter: “Amazon is using this technology to essentially outwit or manoeuvre workers that want to organise.”
Garfield Hylton, a 62-year-old GMB member at BHX4, put it simply: “Workers should be allowed to decide if they wish to choose to unionise. If you’re then monitoring, you’re undermining a process deliberately and willfully.”
Rosa Curling of tech justice non-profit Foxglove called the surveillance “highly intrusive.” Foxglove and Leigh Day are already representing two workers in a case over Amazon’s “click-to-quit” QR codes. Those codes, which let staff cancel their GMB membership, were allegedly stuck on screens, cafeteria tables and toilet doors before the Coventry vote. Curling says they’re now reviewing the Atlas documents for more arguments.
🔍 Why This Matters for AI
Workplace AI usually gets sold as productivity tooling. Amazon already logs every scan and every idle second on the warehouse floor. Atlas goes a step further. It moves from measuring output to forecasting collective behaviour, using sentiment data from both public forums and internal surveys.
That raises three practical issues:
- Survey trust. If feedback surveys feed a union risk model, employees have good reason to stop answering honestly. That damages the data HR teams depend on.
- Legal exposure. The British Columbia Labour Board has already found that Amazon broke several labour laws. Regulators in the UK and EU are paying closer attention to algorithmic management.
- Precedent. Sentiment analysis on scraped social posts is cheap and easy to build. Other employers will be watching how this plays out.
📡 What Comes Next
Expect Foxglove and Leigh Day to try bringing the Atlas material into their existing UK case. Expect GMB to use the Tier 1 label as an organising argument at Coventry and Sutton Coldfield. And if you build or buy workforce analytics, ask a harder question than “can we predict this?” Ask what the prediction will be used for.
The full report, with more worker testimony, is available at the original source.