Amazon Atlas: Internal Documents Detail Union-Risk Scoring

The score runs backwards

Most scores a worker ever sees reward the person being scored. According to documents reported by openDemocracy, Amazon's Atlas inverts that. On the scale described in the company's manager training documents, a warehouse earns a number between 100 and 900 based on how likely its staff are to form a union, stage a sit-in or take other action the documents group together as disruptive labour events. A low number marks the dangerous site. The appendix is explicit about the direction of approval: for leadership, higher scores are "better".

The inputs are what make the number worth reading about. Atlas reportedly draws on hundreds of data points, and the named examples are not limited to shift patterns or local labour-market statistics. They include posts on Reddit forums and employees' responses to internal feedback surveys, alongside broader workplace information and the site's history of industrial action. A feedback survey is something a company asks its staff to fill in so problems can be fixed. In the Atlas appendix, those same survey responses are identified among the information used in forecasting labour activity at the site level. That raises a real question about how employee-generated information is aggregated and protected. But the disclosed material does not establish that Atlas identifies individual respondents, or predicts any individual's personal union activity.

The documents came out of a Canadian labour-board dispute

None of this was published by Amazon. The material was disclosed as part of a legal challenge by Unifor, the Canadian union, and shared with the US non-profit publication Truthout. openDemocracy's investigation, published on 16 October, is built on those documents. The underlying dispute is not a footnote: the British Columbia Labour Board ruled that Amazon had violated multiple local labour laws, following an earlier ruling that the company unlawfully imposed a wage freeze on unionised workers at a warehouse outside Vancouver.

That provenance cuts both ways, and readers should hold both halves. Documents produced by one side of a labour dispute are documents produced by one side of a labour dispute. They are also disclosure from proceedings Amazon lost, describing a system Amazon does not deny exists. What the board's findings do not do is validate every description of Atlas, or establish that the tool itself violated labour law; the ruling was about Amazon's conduct in the dispute, and the board did not independently verify each claim made about the framework. Amazon disputes the characterisation of Atlas — its purpose and its data practices — not its existence.

What Amazon says, and the question it has not answered

Amazon's response has three parts. First, the tool works at site or building level only. Second, the company does not track individual employees' opinions, affiliations or union membership, or any personal data, and Atlas is not used to monitor dissent; its stated purpose is to understand where staff need more support. Third, a framework called Atlas is not currently used in the UK.

The first two parts leave a gap a site-level output does not close. The documents identify staff feedback and Reddit posts among Atlas's inputs, but they do not explain whether those inputs are aggregated or anonymised before scoring, or what safeguards prevent individual identification. Amazon says the framework does not track personal data or individual employees' opinions and affiliations. The disclosed material does not independently establish the processing steps behind that assurance. The third part is narrower than it sounds. "Not currently used in the UK" answers a question about one named framework, in one country, in the present tense. openDemocracy asked whether a similar system runs in Britain under a different name. At publication, Amazon had not answered.

Asked whether Atlas is an appropriate level of worker surveillance, a UK government spokesperson said employers and workers have a shared interest in workplace monitoring technologies being used transparently and in ways that are fair, accountable and reliable, and added: "We expect all employers to respect workers' rights to take part in lawful trade union activity."

Why a tier list matters even if the tool is switched off

Suppose the denial is complete and no version of Atlas touches a British warehouse today. The Coventry and Sutton Coldfield sites still appear, in Amazon's own training material, in a list of 16 sites worldwide classified at the highest level of labour risk, in a document that connects that classification to staffing decisions and actions to mitigate risks to the company's labour structure. Coventry is the site where a recognition ballot was lost by 28 votes in 2024, after the first strikes in Amazon's UK history.

The inclusion of Coventry and Sutton Coldfield in the disclosed Tier 1 list raises legitimate questions about when those classifications were produced, what information informed them, and whether they influenced decisions affecting the sites. The documents do not establish that Atlas is currently operating in Britain, or that the classifications led to particular actions.

The GMB has now put the question formally, in a letter to the chair of Parliament's Business, Innovation, Science and Trade Committee, published by Parliament on 14 October, asking the government to establish whether Atlas, or anything like it, operates in Britain. The next questions are specific: which data categories feed the framework, whether any equivalent system currently assesses British sites, and whether historical Tier 1 classifications informed managerial decisions. Amazon has publicly addressed current UK use of the named Atlas framework, but the reporting leaves the broader deployment and operational questions unresolved.

Sources

[1] openDemocracy, Evie Breese, "Amazon uses AI to monitor union risk in UK warehouses", 16 October 2026 — primary reporting of the disclosed documents: the Atlas appendix, the 100–900 scale and its direction, the five tiers, hundreds of data points including Reddit posts and internal surveys, the 16 Tier 1 sites including BHX4 and EMA4, the Unifor disclosure and BC Labour Board ruling, Coventry strike and ballot history, all Amazon statements including "not currently used in the UK" and the unanswered different-name question, and the UK government spokesperson's statementRead source
[2] UK Parliament, Business, Innovation, Science and Trade Committee — published correspondence: "Letter from the GMB Union regarding Amazon and the use of AI in anti-union activity, 7 October 2026", published 14 October 2026Read source
[3] Computer Weekly, "Amazon uses AI-powered 'risk' system to track union activity globally", October 2026 — independent corroboration of the Atlas framework, tier definitions and the UK sites from the same disclosed documentsRead source
[4] Ground Truth, "Amazon union-risk documents raise questions over Atlas scoring", 18 October 2026 — analytical follow-up on the GMB letter and the site-level versus upstream-data distinction; used for the follow-up framing, not as a source for the document claimsRead source

THE ALFW VERDICT

Internal documents reported by openDemocracy describe an Amazon framework, Atlas, that combines workplace information, employee feedback and public online material into a single site-level score for perceived union-organising risk — a scale on which lower means riskier, and on which two British warehouses appear among the highest-risk sites in the disclosed material. None of that establishes that Atlas currently operates in the UK, that it identifies individual workers, or that any score caused a particular managerial decision; Amazon denies individual surveillance and says the named framework is not currently used in Britain. The unresolved question is a policy one: whether forecasting lawful collective action out of workplace data can be done transparently, with adequate protections for workers' rights. That cannot be answered by the score itself. It requires clarity about what data goes in, what safeguards apply, where the framework runs, and which decisions it informs.