Treat the hidden workforce that produces AI — content moderators, data labellers, gig drivers whose data trains routing systems, algorithmically managed warehouse and platform workers — as the labour movement organises any other industrial workforce: build associations and unions, contest contracts, win legal status, and confront the named corporate buyer at the top of the chain. The point of leverage is that this labour is structurally necessary to the AI product, so a credible threat to withdraw it is a credible threat to the product.
An actor chooses this strategy because policy advocacy alone cannot move the conditions under which AI is actually built — the supply chain runs through low-wage jurisdictions deliberately chosen for distance from regulation. A workers' organisation can do what an NGO cannot: extract a binding agreement that changes the conditions of the work, force a buyer to acknowledge an employment relationship it has structured itself out of, and produce first-person testimony of harm with standing in court.
It trades off the slowness of building real organisations in hostile labour environments — many of the workforces are precarious, often outsourced through multi-tier vendors, in countries where union recognition is itself a fight. Wins are concentrated where labour law already protects organising; transnational solidarity is required to close the supply-chain loop.
Verdict — good strategy, narrow but cumulative
Strong on its primary claim — that the supply chain is real, organisable, and legally addressable — and so far thin on its secondary claim, that organising produces material change in the conditions of the work. The four effects: entries above are what the strategy returns to the movement. The Kenyan Court of Appeal's 20 September 2024 ruling is the strategy's most consequential single output: an African appellate finding that the US platform parent can be sued in the African jurisdiction where its outsourced AI-supply-chain work is performed, and the procedural unlock without which every other Nairobi-cluster case would have died at the jurisdiction stage. The 1 May 2023 founding of the African Content Moderators Union is the strategy's signature recognitional milestone — a cross-platform union of workers from Sama, Majorel, and Teleperformance who review Facebook, YouTube, TikTok, and ChatGPT — and is matched a year and a half later by the Data Labellers Association launch on 13 February 2025, which extends the form from content moderation into the data-labelling layer the frontier-model RLHF pipeline runs on. The Amsterdam Court of Appeal's 4 April 2023 GDPR Article 22 ruling in the Uber/Ola case carries the strategy's logic from the African content-moderation outsourcing chain into the European gig-economy algorithmic-management chain, and establishes that worker-side litigation under data-protection law can force platforms to disclose how their algorithmic systems actually allocate work, set pay, and dismiss workers — a transparency win on the legal substrate that organising and bargaining can stand on.
The strategy is bad in two distinct ways. First, by the pace of return. The two flagship Kenyan dockets (Motaung's Petition E071 of 2022; the 185-moderators petition E052 of 2023) have been on the Employment and Labour Relations Court's record for three and four years respectively, the Court of Appeal jurisdictional ruling clears the path to trial but trial has not started, and the parliamentary petition has produced no statutory output in three years. The strategy returns evidence, public identity, and law before it returns wages or contracts; an adopter that needs a near-term material return will not get one. Second, by the platforms' available counter-move. The supply-chain worker's leverage — the threat to withdraw structurally necessary labour — is degraded by the platform's freedom to close the shop floor: Sama's January 2023 mass redundancies, the OpenAI/Sama contract terminated eight months ahead of schedule in February 2022, Uber's open non-compliance with an Amsterdam court order even at €4,000 per day. The strategy works against a buyer that needs the labour in this jurisdiction; against a buyer that can move the labour, it produces an evidentiary record of what happened rather than a change in what happens next.
The strategy's adopter mix matters more here than on most strategies. An adopter that runs only the worker-organising arm — without the empirical audit evidence base that gave the Nairobi cluster its TIME, Truthout, and Privacy International record, without the strategic litigation that wins the procedural unlocks, without the survivor-led testimony anchor (Motaung, Mathenge, Okinyi, Kinyua) that gives every public-facing moment a named worker — does not produce the stack of recognitional and judicial wins the Nairobi cluster has produced. The strategy works inside that stack; it loses badly as the lone arm.
Ecology
This strategy is paired with strategic litigation against algorithmic state decisions as the legal sibling of the same coalition — in Nairobi, the Foxglove / Nzili & Sumbi cluster runs both arms (the union founding plus the petitions docket) and shares plaintiffs across them (Motaung is the lead worker-petitioner in the Motaung suit and the namesake of the Alliance that founded the union; Mathenge is a founding ACMU figure and a lead parliamentary petitioner). It is fed by empirical audit and expose — TIME's January 2023 investigation into how Sama's Kenyan workforce built ChatGPT's safety layer, Worker Info Exchange's Managed by Bots report on seven major gig platforms, Privacy International's long-read on gig-economy surveillance — without that investigative substrate the worker-organising arm has neither the testimony anchor nor the international press attention that makes a 150-worker union founding a Reuters story. It uses survivor-led testimony as its public-facing register: Daniel Motaung's "I never thought, when I started the Alliance in 2019" address at the founding summit, Mathenge's "destroyed me completely" testimony, Joan Kinyua's "AI does not exist on its own. Behind every algorithm, every dataset, and every technological advancement, there is invisible labor" — the strategy's content travels on named workers, not on principles. It feeds coalition lobbying of binding regional regulation as the worker-side evidentiary input — the Kenyan parliamentary petition, the DLA's submissions to Kenya's Business Laws (Amendment) Bill 2024, the WIE/ADCU evidence that anchors the Dutch DPA's €290 million Uber fine — and in the EU AI Act trilogue period the content-moderation labour question entered the coalition record as a high-risk-system fundamental-rights argument that the coalition could not have made without the workers' own organisation behind it.
This strategy parallels creator-class collective bargaining on generative AI (WGA, SAG-AFTRA, the Concept Art Association) — the form is the same (organise the workers whose labour the AI replaces or appropriates; bargain with the named corporate buyer), but the position in the supply chain is different (creators sit upstream of the model as the source of training data; supply-chain workers sit inside the pipeline as the source of safety and routing labour), and the two adopter populations rarely overlap. It also parallels tech worker refusal inside AI labs — both strategies put workers at the centre of the AI-good question — but the lab-worker strategy depends on insider professional power against a lab employer, while the supply-chain strategy depends on outsider supply-chain power against an opaque outsourcing buyer. The strongest competing strategy is the platform's own counter-strategy: move the work. Where the platform can shift content moderation from Nairobi to Cape Town, or RLHF labelling from Sama to a new vendor, the supply-chain worker's leverage decays toward zero. The strategy's geographic concentration (the Nairobi cluster; the Amsterdam/UK gig-driver cluster) is a measure of where the counter-strategy has not yet been deployed.