Amazon uses AI-powered 'risk' system to track union activity globally | Computer Weekly

TEXT START: Internal documents disclosed during a Canadian labour dispute show Amazon is using an artificial intelligence-powered risk-scoring system named Atlas to track union organising among its workforce, while training managers on how to handle what it perceives as the ‘threat’ of union activity

The Dissection

The article documents a labor-control apparatus. Atlas converts turnover, incidents, worker sentiment, wages, appeals, nearby union activity, internal and external discussions, and manager-reported events into site risk tiers and severity scores.

Amazon’s defense is semantic containment: no individual monitoring, only aggregated site-level feedback. That distinction does not answer the functional question. A system designed to predict where collective organizing may emerge and classify union cards, protests, sickouts, and work stoppages as high-severity events is tracking labor resistance, even if it does not identify every worker individually.

The article exposes the mechanism but limits its frame to privacy, labor law, union rights, and corporate compliance. It reports the instrument without fully tracing how AI strengthens capital’s ability to preempt worker coordination while the old employment order decays.

The Core Fallacy

The central fallacy is treating this as a bad-surveillance episode that can be solved through disclosure, data-protection rules, manager training, or expanded union access. Those are lag defenses. They may constrain the method temporarily; they do not alter the competitive incentive to make labor unrest legible, rankable, and preventable at scale.

Under DT logic, AI is not merely watching workers. It is converting collective resistance into an operational risk variable. Unions can contest that machinery, but they cannot restore the mass employment → wage → consumption circuit once human labor becomes structurally less necessary.

Hidden Assumptions

• Aggregated data is presumed less threatening than individual surveillance, although collective organizing is the target.
• Legal compliance is treated as capable of preserving meaningful worker power.
• Transparency and oversight are presumed able to neutralize the competitive pressure to anticipate and disrupt coordination.
• Unionization is framed as a discrete labor-relations dispute rather than part of the wider struggle over productive participation.
• The current name of the system is treated as significant; rebranding does not change its function.
• Worker bargaining power is assumed to remain durable enough to discipline capital, even as AI expands management’s ability to substitute, monitor, and preempt.

Social Function

Partial truth with a transition-management and ideological-anesthetic effect.

The article punctures Amazon’s corporate denial and reveals genuine control infrastructure. But by centering the question of whether Atlas legally qualifies as surveillance, it channels the response toward regulators, litigation, and procedural safeguards. It exposes the weapon while leaving the battlefield’s underlying structure largely intact: AI is presented as a governance controversy rather than as a force determining who remains economically necessary.

The Verdict

Atlas is a corporate immune system for the late labor market: it detects collective symptoms before they become coordinated resistance. Amazon’s denial is linguistic camouflage. “No individual surveillance” does not erase site-level prediction of union activity.

The article does not by itself prove that P1–P3 are complete; it does not establish AI dominance across all cognitive work or majority exclusion from productive labor. It does establish something narrower and strategically important: AI is already being used as a lag defense to discipline and preempt workers while the postwar employment order weakens.

The report identifies the smoke accurately. Its reformist frame understates the fire.