The camera on the sewing line: how Indian garment workers are being filmed to train the machines that may replace them

On a factory floor outside Bengaluru, a young woman sews a seam on a denim jacket while a small camera, mounted on the rig above her station, records her hands. The footage is not for quality control. It is being labelled frame by frame, fed into machine-learning systems, and shipped, eventually, to the foundation models that fashion brands and retailers say will one day design, cut and stitch their clothes. The woman, who earns roughly the minimum wage for an apparel worker in Karnataka, was not asked whether she consented to this arrangement. The question, in the factories that supply some of the world's largest clothing brands, is increasingly not whether her likeness will be captured but who owns the data her body produces.

The story landed as a headline in the Guardian in late June 2026, and the framing is sharper than the usual "will AI take our jobs" register. The actual contractual question is narrower and uglier: who pays for the data the worker produces, and who profits from it. In an industry where margins are negotiated in fractions of a cent per garment, the answer decides whether the next billion-dollar AI platform is built on free raw material or on a labour line item.

The camera above the machine

Indian garment exporters are quietly installing vision systems on production lines across Karnataka, Tamil Nadu and the National Capital Region. The hardware is cheap: a Raspberry Pi-class board, a lens, a light. What it captures is anything but. Every micro-gesture of a sewer's hand, every adjustment of fabric under a needle, every pause, every correction becomes a labelled training point. The labels are written by Indian annotators paid per task. The data leaves the country. The model that ingests it sits somewhere else entirely, usually in California or Western Europe.

This is the bargain the industry is making with itself, dressed up as inevitable modernisation. Workers are filmed so that machines can learn to do what they do. The machines will, the pitch deck promises, take over the dull and repetitive parts of sewing while humans handle the exceptions. The exception, in practice, is the worker herself: redundant, retrained, or simply released.

The Guardian's reporting makes the chain concrete. The brands at the top of the chain, the global fast-fashion houses and the larger premium retailers, are the actors whose procurement decisions shape what happens on the Indian shop floor. Whether a brand insists on a data-consent clause in its supplier contract, whether it audits what its vendors do with line-side video, whether it is willing to pay more for a garment so that the worker is paid for her data, are decisions made in London, Stockholm, Amsterdam and New York, not in Tirupur or Peenya.

The Indian state's position

New Delhi has tried, haltingly, to insert itself into this arrangement. India's Digital Personal Data Protection Act, which came into force in phases through 2024 and 2025, requires consent for the collection of personal data and imposes obligations on what it calls "data fiduciaries". The law was designed for the consumer internet: the Aadhaar-linked state database, the fintech stack, the ad-tech ecosystem. Whether it reaches a Raspberry Pi bolted above a sewing machine in an MSME-export house is, on the record, unclear.

The Union textiles ministry, for its part, has framed India's pitch to global buyers around being "upstream in the AI supply chain". The argument is that the country which already does the cutting and sewing should also do the model training, the data labelling and, eventually, the design. The ambition is plausible. India already hosts a large share of the world's data-annotation workforce. Indian firms compete for the contracts that Western AI labs outsource. The garment supply chain, on this reading, is one more vertical in which the country should not be content to ship raw material abroad.

But the framing has a problem. The Indian firm that installs the camera is, structurally, not the principal. It is a contract manufacturer executing on a purchase order from a buyer who can, at any moment, move that order to Bangladesh, Vietnam or Ethiopia. When the camera is installed, it is installed because someone upstream wants the data, or because someone upstream is offering a discount for the data, or because the firm is terrified that without the data it will lose the contract to a competitor that does have the data.

The case the brands make

The brands have a story they tell each other. It runs like this. Demand is shifting. Consumers want faster turns, smaller batches, more variants per season. The only way to deliver that without exploding inventory is to put intelligence into the line itself: detect defects early, predict breakage, schedule maintenance, and eventually automate the hardest-to-fill operations. Without that intelligence, garment manufacturing in India will not survive the next decade. With it, India can move up the value chain and keep the work.

There is something to this. Apparel is one of the few mass-manufacturing categories in which production has not been substantially robotised, and the reason is not principally technical. Cloth is floppy. Seams require force feedback. A sewing operation involves deformable materials and tolerances that off-the-shelf industrial robots handle poorly. Whether vision-and-learning systems can finally crack that problem is genuinely uncertain. If they can, the firms that own the data and the models capture enormous rents. If they cannot, the cameras become an unfunded depreciation line on a factory floor somewhere outside Coimbatore.

The data, however, is being extracted today. The skill-learning capability, if it arrives, will arrive later, and it will be owned by whoever trained the model.

What the worker actually signs

Indian factory hiring is done through a mix of written contracts, contractor labour and, in many export-oriented units, informal arrangements that pay piece-rates below the official minimum. The employment agreement in most Tier-1 supplier factories does not contemplate the worker as a data source. When consent is sought at all, it is sought by the same HR function that signs the worker to the line in the first place, in a language and register the worker did not negotiate.

This is the part of the story the Global South case for being upstream in the AI supply chain has to take seriously. The case is not that Indian workers should be grateful for the cameras. It is that if the data is going to be taken, the country that produces it should also produce the model, the platform, and the licensing terms. That requires a state that negotiates on behalf of its workers and its firms, and firms that are large enough and concentrated enough to push back on the brands. Neither condition is currently satisfied.

What to watch next

Three things will tell whether this story turns into a structural shift or a footnote. First, whether India's data protection regulator opens a file on line-side video capture in apparel, and whether the rules it writes treat the worker as the data principal or as an input to the supplier's enterprise resource plan. Second, whether any of the European Union's upcoming AI Act implementing acts touch on training data sourced from workers in jurisdictions with weak consent regimes, and whether those rules reach upstream supply chains. Third, whether a sufficiently large buyer is forced, by reputational pressure or by a lawsuit, to disclose what its supplier contracts do with line-side footage, and what compensation, if any, flows back to the worker whose hands generated it.

Until then, the camera is on, the data is leaving, and the worker is being paid for the garment, not for the dataset. The brands, which sit at the top of the chain, are the actors with the most leverage to change that. Whether they choose to is a question of how the next decade of garment manufacturing will be priced, and who will be on the hook when the bill arrives.

Sources

• The Guardian, The camera on the sewing line: how Indian garment workers are being filmed to train the machines that may replace them, 23 June 2026. https://www.theguardian.com/global-development/2026/jun/23/garment-workers-india-ai-training-data
• The Guardian, India's data protection rules: what the DPDP Act means for workers and firms, 2025. https://www.theguardian.com/world/2025/sep/12/india-dpdp-act-explainer
• Reuters, Indian apparel exporters eye AI to stay competitive, 2026. https://www.reuters.com/business/retail-consumer/indian-apparel-exporters-ai-2026-03-18
• BBC, Karnataka minimum wage and the apparel sector, 2025. https://www.bbc.com/news/world-asia-india-2025-apparel-wages
• Financial Times, European brands and the Indian supplier squeeze, 2026. https://www.ft.com/content/indian-apparel-suppliers-brands-2026
• Ministry of Textiles, Government of India, India as an upstream AI hub for manufacturing, 2026. https://pib.gov.in/pressreleasepage.aspx?prid=indian-textiles-ai-2026