Concerns Over AI Training Data Collection from Gig Workers
The Ministry of Electronics and Information Technology has expressed worries regarding the gathering of first-person video along with sensor data from gig workers to enhance physical AI models. This scrutiny comes in light of Pronto’s initiative to record videos within homes, while competing platforms Snabbit and Urban Company have opted against implementing similar methods.
As reported by ET, high-ranking officials from the ministry convened with representatives from home services platforms such as Snabbit, Pronto, and Urban Company, as well as technology and robotics firms on Thursday. The discussion focused on issues surrounding workers’ rights, consumer privacy, and data sovereignty. It is essential to note that their participation in these consultations does not imply that all three platforms are currently accumulating household footage for AI training.
Details of the Ministry’s Meeting
The meeting, led by MeitY secretary S Krishnan, aimed to gather insights on the standard operating procedures for collecting, processing, and storing first-person video recordings. The ministry also inquired about measures the companies have in place to secure workers’ consent and to safeguard the privacy of household members and others featured in the recordings.
Officials questioned whether the gathered data was being sold, shared, or transferred to AI developers located outside India. The ministry called for transparency regarding data-sharing agreements, vendor contracts, retention policies, as well as the type and volume of information being collected.
Pronto’s In-Home Video Recording Initiative
Among the three home services platforms, Pronto has taken the step to implement in-home video recording through its Verified service. Conversely, Snabbit and Urban Company have confirmed they will not engage in this practice. As reported by ET, Pronto has been requested to provide a comprehensive account of its data collection methods and standard operating procedures.
During the meeting, some participants raised concerns about the implications of collecting data in India and subsequently selling it to overseas physical AI developers, suggesting it does not benefit the country. Robotics startups highlighted the need for India to prioritise the development of its own physical AI capabilities over supplying training data to foreign companies.
Industry Statements on Data Practices
Pronto stated its participation in the MeitY discussion was part of a widespread industry dialogue on responsible data practices. The company asserted that the data generated through its customer booking process is obtained with customer consent and can contribute to downstream AI initiatives. They also noted that their existing operations comply with relevant regulations and that they will continue collaborating with the ministry as policies evolve.
Investor Interest in Physical AI in India
Pronto’s data collection initiative follows the startup’s fundraising success, having secured $20 million from US tech investor Lachy Groom, who co-founded the AI startup Physical Intelligence. Anjali Sardana, the founder of Pronto, has launched a new physical AI venture called Provision Labs, collaborating with developers focused on creating AI models for robotics and similar real-world applications.
Interest from investors in India’s physical AI sector has noticeably increased. According to a report from Startup Superb, ventures within this area raised approximately $155 million from 31 deals by July 2026, a rise from $130 million from 25 deals in 2025.
Building Infrastructure for Physical AI
A distinct group of companies is focused on constructing the frameworks necessary for data collection, annotation, and training for physical AI systems. Human Archive is working on developing datasets that capture human sensorimotor activities for robotic use, while Neo Cambrian has introduced an India-centric robotics data factory. Other active companies in this sector include iMerit, Human Stryde, Luel, Awign, Build, Centific, Humyn Labs, Aura ML, Build AI, FPV Labs, and Northstar.
Importance of Real-World Data for AI
Physical AI models depend on genuine data that reflects human actions, objects, environments, and interactions. A significant portion of this data needs to be collected specifically for robotics and embodied AI applications instead of being sourced from existing datasets available online.
Legal experts cited by ET have pointed out that video recordings, facial images, location data, and sensor information could be classified as personal data when they assist in identifying or can reasonably trace back to a worker. Under the Digital Personal Data Protection Act, platforms are obliged to provide clear notifications concerning data collection and processing purposes, rights of workers, and mechanisms for lodging grievances.
This development follows previous dialogues among startups working within the realm of physical AI, aimed at establishing common standards for worker compensation, wellbeing, working hours, safety, and legal compliance.