On a Tuesday evening in early July, Anushka Ahuja, a dentist in Gurgaon, India, opens the door for the house help she booked ten minutes ago on Pronto, an app that provides instant professional cleaning services like sweeping, mopping and laundry across 11 cities in India. A woman clad in a crisp uniform greets her with a polite smile, then asks: Can she strap a small camera onto her head to record her work?

Ahuja isn’t surprised to see the camera. In fact, she paid an extra nine rupees (about nine cents) so her cleaner would record everything from her elbows to her fingertips. Now she’ll receive a video recording a few hours later, giving her peace of mind that her house was cleaned properly and nothing was stolen or broken in her absence.

Despite disclosures in the consent form she signed, Ahuja initially didn’t realize that there’s another purpose to the camera footage: AI training data. After a cleaning session, the video is stripped of personally identifiable information like personal documents and people’s faces and then sold to American AI companies.

Bangalore-based Pronto launched the video recording service, called Pronto Verified, in May, stirring widespread backlash in India and government scrutiny over privacy concerns. Customers consent to the recording and workers are paid double for Pronto Verified bookings, CEO Anjali Sardana says.

“Trillions of dollars are going into AI capex every year. All the money in the world is getting invested in AI, but your blue-collar worker in India is not capturing any of that,” Sardana, 24, tells Forbes. “And there is an opportunity to actually redistribute that capital.”

Ahuja says she doesn’t mind that footage of the inside of her home will end up in an AI model’s training dataset. “Everyone is just collecting data… I think it's just everywhere, so it's fine,” she says. “It’s beyond our control now.”

Demand for Pronto’s video footage has exploded in the past 12 months as robotics becomes the new hot category in AI, says Ajay Agarwal, a partner at Bain Capital Ventures, which invested $2 million into the startup when it was founded in February 2025. Unlike large language models, which were trained using data scraped from all corners of the internet, robotics startups require data that’s not as readily available, including videos of people interacting with physical spaces. “The thing we're able to offer the AI labs is incredible diversity of environments with 50,000 cleanings a day, a range of humans performing different cleaning tasks and that's a scale that no one else has, frankly,” Agarwal says. Pronto declined to share which AI startups are using its data.

“The thesis behind Pronto is just that broken labor markets should not exist.” Anjali Sardana, CEO of Pronto

“The thesis behind Pronto is just that broken labor markets should not exist.”

The most immediate application for Pronto’s data: building a........

© Forbes