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

On 24 June 2026, the Guardian reported a question that has begun to settle over India's garment belt like humidity: who pays the workers who teach the machines that are meant to replace them? In factories around the southern city of Bengaluru and across Tamil Nadu, supervisors are attaching head-mounted cameras to the heads of sewing-line operators, recorders the size of a small torch clipped above the brow. Each clip captures the micro-movements of a single task — the way a thumb feeds a sleeve under a needle, the half-second a hand reaches for a thread-cutter, the precise angle at which a finished shirt is folded for inspection. The footage is being shipped, often the same day, to artificial-intelligence labs in India and abroad. It is, the workers are told, the future of the factory. It will also, the workers suspect, eventually be the end of theirs.

What is unfolding in these sheds is a quieter, slower version of a confrontation that has been headline news in other industries: the capture of human skill, in granular form, by systems that need not sleep, file grievances, or take the bus home. Garment work has long been sold to the world as a rung out of rural poverty, an export-led step into industrial modernity, a source of dignified employment for tens of millions of women in particular. It is now being asked to perform a second function — to record itself into obsolescence. The question is not whether Indian garment exports will survive the AI transition. By most measures they will, and grow. The question is who in that chain captures the value of the data the workers generate, and on what terms.

A new line on the production sheet: 'data work'

The Guardian's reporting describes workers like Lalita, a name changed for her protection, who was given a head-mounted camera and told that her shifts would be filmed. The novelty, she told the paper, wore off in a day. Within a week the cameras were routine; within a month the workers were submitting to them the way they had long submitted to piece-rate targets, attendance logs and exit searches. The pay for the filming, where it is itemised at all, is small — a few rupees per usable clip, sometimes a flat monthly top-up, rarely a renegotiated wage. The clips themselves are not owned by the worker. The footage leaves the factory, often before the worker has removed her badge, and is uploaded to systems she has never seen and cannot audit.

The economic logic is straightforward. A skilled sewing-line operator takes months to train and seconds to lose; a stitch produced by a vision-based robotic system is, in theory, replicable indefinitely. The intellectual property generated by an Indian worker's hand is, in the legal sense, almost always the property of the factory or its foreign buyer, not of the worker who created it. Indian copyright law does not, in general, recognise a performer's right in the motor patterns of their own work; it recognises the employer's right in the recording. The data is treated as a by-product of employment rather than as a contribution to a new and extraordinarily valuable asset.

The counter-narrative: modernisation, demand, dignity

The industry's response, in the limited public comments available, is that the cameras are voluntary, that the work is no different in kind from a supervisor's shadow on the line, and that the technology is the only way Indian garment exports can keep pace with buyers in Europe and North America who are themselves under pressure to automate. There is some truth in each of these claims, and they deserve to be taken seriously. The wage gap between India and the garment hubs of Bangladesh, Vietnam and (increasingly) Ethiopia is narrowing, but productivity has not caught up. The global apparel market is in a slow squeeze: input costs are up, fast-fashion margins are down, and the brands that buy from Indian factories are themselves being asked by their shareholders to defend margins by automating everything they can. Refusing to record workers' hands is, in this framing, a route to losing the order book.

There is also a Global South case to be made for being upstream in the AI supply chain rather than downstream of it. If the data captured in Indian factories is what trains the vision systems that will eventually run garment production in dozens of countries, Indian firms — and, in principle, Indian workers — have a claim to sit closer to the value. Some Indian industrial-policy voices, including in the wider conversation around digital public infrastructure, have argued that the country's data should be treated as a strategic asset, the way its rare-earth deposits or its generic-pharmaceutical capacity are. In that reading, the camera on the worker's brow is not an indignity but a deposit in a national AI endowment.

The structural frame

Both readings are partly right, and that is exactly the problem. The pattern unfolding in these factories is a familiar one in late-industrial development: a high-skill, low-wage workforce generates a public good — clean air, raw data, a trained hand — and the rent from that good is captured elsewhere. The garment industry has been here before. The development economics of the sector has been debated for decades: whether low-wage manufacturing is a ladder or a trap, whether export-led industrialisation is sovereign or subcontracted, whether the workers who built the Bangladesh garment industry captured a fair share of the wealth they generated. The current moment adds a new variable. The asset being produced is not just a T-shirt; it is the model that will eventually make the T-shirt without human hands. Once the model is trained, the workers who trained it are not just low-paid; they are, in a strict sense, redundant by design.

The question this poses is not technological. The cameras are crude, the AI is imperfect, and full automation of a garment line remains stubbornly harder than automation of a warehouse. The question is contractual. Are the workers who generate the data being paid for it? Are they being asked to consent to it? Is there a mechanism, in Indian law or in the buying brands' supplier codes, that treats a worker's motor patterns as a contribution to intellectual property rather than as a free input? The Guardian's reporting suggests that, in most of the factories it examined, the answer is no.

What remains uncertain

The scale of the practice is unclear. The Guardian's report draws on a small number of named factories and a larger number of anonymous workers, and it does not claim to estimate how many of India's several million garment workers are now being filmed. The companies operating the programmes are not all named; some appear to be Indian AI startups, others to be foreign buyers with in-house automation teams. The Indian federal government has not, as of the report's publication, issued guidance on the data rights of workers whose movements are recorded in the course of their employment. State-level labour ministries in Karnataka and Tamil Nadu, where much of the filming is concentrated, have not announced investigations. The brands at the top of the supply chain — the European and North American retailers whose orders pull the whole system along — have not been asked, on the record, to explain what they are doing with the footage their suppliers collect.

What can be said with confidence is narrower. Workers are being filmed in Indian garment factories. The footage is being used to train AI systems. The workers are not, in most cases, being paid a premium for the filming or given any ownership of the resulting data. The buyers and the factories have, so far, declined to publish the contracts that govern this trade. The question the headline put to its readers — who is going to pay us when we're replaced by robots? — does not yet have an answer from anyone in a position to give one.

Stakes

If the present trajectory continues, the value of garment work in India will be repriced downward long before it disappears. The workers who train the systems will earn piece-rates while their data earns venture-scale returns. The brands that buy the resulting garments will capture the productivity gains. The Indian state will see jobs lost and a strategic asset — a national corpus of skilled-motion data — sold at no defined price. None of that is inevitable. Indian labour law, Indian industrial policy and Indian diplomatic posture in the wider conversation about AI governance could, in principle, treat worker-generated data as a category that exists in law, with corresponding rights to consent, compensation and, where appropriate, co-ownership. Whether they will is the open question that the cameras on the sewing line have, suddenly and uncomfortably, put on the table.

Desk note: Monexus framed this piece around the contractual question the Guardian's headline raises — who pays for the data the worker produces — rather than around the more familiar 'will AI take our jobs' framing. The Global South case for being upstream in the AI supply chain is taken seriously, and the brands at the top of the chain are named as the actors whose conduct most needs scrutiny, since the Indian state and the Indian firms are, in different ways, responding to pressure that originates with them.