Hundreds of thousands of people around the world are now paid to record their own activities on camera, and the footage is sold to train the AI that controls robots, the BBC reported on October 23, 2026. For a barber in Bengaluru, that means a camera strapped to his head for five or six hours a day while he cuts hair, framed so only his hands and fingers show.
The 22-year-old has cut hair for four years. “I never imagined my razor techniques would be useful to a global tech company,” he told the BBC. His footage goes to a local firm that may have hundreds of people filming their work. The firm sells it to technology companies and research organizations building so-called physical AI, software that controls robots in the real world.
From one head camera to six
Rashdan Technologies is one firm in this trade. Its co-founder and CEO, Shahil Baig, told the BBC that every robotics company and frontier AI lab wants a specific kind of footage: industrial tasks, agriculture, household chores and small-business work such as pottery, garment-making, embroidery and candle-making. “We go wherever the task is,” he said.
The company pays $5 to $10 for eight to 10 hours of recorded work, roughly 50 cents to $1.25 an hour by our arithmetic. Baig says Rashdan records “hands, not faces,” blurs faces when a rig captures more of the scene, and passes on nothing that identifies the worker.
The rigs are getting bigger. “Now clients are asking for four to six cameras rigged around a person’s head to cover the entire scene, not just the hands,” Baig said. Rashdan pays more for that tier, he added, because it asks more of the people wearing it. The founders the BBC interviewed expect dozens more competitors to appear in India this year.
Footage the internet doesn’t have
Manish Agarwal, founder of Humyn Labs, a data company with bases in Bengaluru and San Francisco, told the BBC that robots built for the real world cannot be trained inside computer simulations alone. “There is no existing internet archive of embodied human activity,” he said.
Humyn Labs has 96,000 data collectors in Latin America and Southeast Asia, according to the BBC. Most of the data is video, but some collectors wear hardware on their hands and fingers to capture physical contact. In Agarwal’s example, a robot has to learn how hard to grip a plastic bottle so it doesn’t crumple, and how that differs from a steel glass. On its website, the company calls itself “the research & data engine for Physical AI” and lists a corpus of more than 200,000 hours of first-person video for humanoid systems.
Agarwal also told the BBC that a robot meant for American homes or factories needs footage from places that are a proxy for them, and that “residential houses in India may not be the right context.”
Karur, a town on camera
NPR reported from Karur, a small city in southern India, in two stories published October 14 and October 15. Developers need “millions of hours” of what the industry calls egocentric data, first-person footage of human work, which “can’t be scraped off the internet,” NPR reported. Prathosh A.P., a professor at the Indian Institute of Science in Bengaluru, told NPR to think of today’s AI systems as “brains without a body” that now need motor skills.
Objectways, the local data firm, appears to be one of the town’s biggest white-collar employers and occupies at least four building blocks, according to NPR. One floor is fitted out as a three-bedroom home where people film themselves chopping onions or wiping tables and cupboards. A senior employee, Rajeshkumar Ramaswamy, said the company pays thousands of people to record themselves, from the Philippines to Nigeria.
Most clients for household footage are in the United States, so Objectways also pays Americans to film their housework. Ramaswamy explained why with a plate: many Indians eat from stainless steel, which survives a robot that grips too hard, while the same grip would smash the ceramic plates many Americans use.
Two or three hours of chores a day for $120 a month
A science teacher in Guntakal told the BBC she has spent seven months filming herself scrubbing, cooking, sweeping and folding laundry at home for “two to three dollars per hour.” Recording three to five hours a day, she said, makes “a good amount in hand” on top of her teaching.
In Karur, a 24-year-old woman films two or three hours of chores a day for about $2.60 an hour and makes around $120 a month, NPR reported. Her mother, who had wanted her to focus on her studies, had to show her how to do the chores.
A Bengaluru tailor with 10 years in the trade told the BBC he now wears a head camera at work to send more money to his family in Bihar. “This is survival,” he said. “It doesn’t cost me anything. I’m already doing the stitching, already doing the cutting.”
At the Texworth textile factory in Karur, NPR described three men handing GoPros and smart glasses to workers. One of those workers, a 40-year-old woman who packs aprons, has worked there six days a week for 10 years and earns $90 a month. She says the company never told her the camera was training robots. “Nobody told me anything about it,” she told NPR.
A 42-year-old colleague is one of the few workers there paid about $2.60 an hour to record herself, and she knows what the footage is for. She expects a robot to take her job one day and is thinking about taking up farming when it does. Until then, she told NPR, she is proud to teach it to do a good job.
Where the reporting stops
The “hundreds of thousands” figure is the BBC’s own, given without a method. The 96,000 is Humyn Labs’ count of its own collectors, and Rashdan’s account of consent and blurred faces comes from the company itself.
Consent is uneven. NPR reported that many workers, like the Texworth packer, appear to be made to record without explanation. Astha Kapoor, co-founder of the Aapti Institute, which studies how technology affects people, told NPR: “Call it what it is, which is exploitation of labor.”
Bino Paul, a senior professor at the Tata Institute of Social Sciences in Mumbai, told the BBC it is too early to tell whether these workers are collecting data for machines that will one day do their jobs. “By the time workers realise what they have participated in, things may have gone too far,” he said.
NPR reported that robots are unlikely to be deployed at scale in India soon, because they are expensive and draw a lot of power, which is unreliable and costly in parts of the country. “Humans are cheaper to run than robots,” Kapoor said. Objectways CEO Ravi Shankar agreed that robots would cost more than human labor. His Indian clients, he told NPR, want robots for jobs that are hard to staff, remote or physically grueling: tapping rubber trees, dusting solar panels in rural areas, destemming chiles.
What it means
The bottleneck in humanoid robotics is now footage of real hands, and a labor market has formed around it. According to AI industry experts cited by NPR, thousands of people across India found this work over the past year. AI is already changing other kinds of work: in Anthropic’s hiring data, juniors fell from 27% to 9% of technical hires; Barclays named AI among the reasons for about 5,000 job cuts this year; and Waymo’s robotaxis have driven 271 million miles with nobody in the driver’s seat.
For the people filming, the pay comes now; whether the footage later costs them their jobs is, as Paul told the BBC, too early to tell. The 24-year-old in Karur makes about $120 a month from a few hours of chores a day, more than the $90 the apron packer earns working six days a week. Agarwal says India’s current role in the trade is “largely as a supplier of raw material,” and expects the value of collection work to erode as more firms compete for contracts. He wants India to move up the value chain, to cleaning, structuring and validating the data.
The question now is whether the work on camera, a barber’s razor technique or a tailor’s cutting, comes back as robots doing the same tasks, and at what price. In India, the pace will depend on the running costs Kapoor and Shankar describe and on a power supply NPR calls unreliable in parts of the country. For now, two signals in the BBC’s reporting are worth tracking: clients asking for four to six cameras per worker, and the dozens of new collection firms the founders expect in India this year.
Sources
- The Indian workers filming their work to train AI, BBC News, October 23, 2026
- These workers help train robots who could one day steal their jobs, NPR, October 14, 2026
- To train robots, AI companies are paying people in India to film themselves doing chores, NPR, October 15, 2026
- Humyn Labs | AI Training Data for Robotics & Physical AI, Humyn Labs, accessed October 24, 2026
Header image: chart by MACCUS, data from BBC News (October 23, 2026) and NPR (October 14 and 15, 2026)
