India's MeitY Probes Gig Platforms Over In-Home AI Data Harvesting

India's electronics and IT ministry has summoned home services platforms to explain why they are recording first-person video inside customers' homes to train physical AI models — and whether that data is being sold to overseas developers.

The Ministry of Electronics and Information Technology (MeitY) met representatives of Snabbit, Pronto and Urban Company, alongside technology and robotics companies, on Thursday, according to an ET report. The meeting, chaired by MeitY secretary S Krishnan, covered three pressure points: workers' rights, consumer privacy and data sovereignty.

Only one of the three platforms actually records inside homes. Pronto runs in-home video capture through its Verified service. Snabbit and Urban Company have both ruled out adopting the practice. That split left Pronto alone in the regulatory spotlight — the ministry has asked the startup to submit a detailed representation on its data collection practices and standard operating procedures.

What did the ministry want to know?

The ministry demanded specifics, not assurances. Officials sought details on standard operating procedures for collecting, processing and storing first-person video recordings, and asked what safeguards exist to obtain workers' consent and protect the privacy of household members and other people who appear in the footage.

The questions did not stop at capture. MeitY pressed companies on whether collected data was being sold, shared or transferred to AI developers outside India. It also requested disclosures on:

• Data-sharing agreements
• Vendor contracts
• Retention policies
• The nature and scale of information being collected

Legal experts quoted by ET noted that video recordings, facial images, location traces and sensor data could qualify as personal data when they identify, or can reasonably be linked to, a worker. Under the Digital Personal Data Protection Act, platforms must provide clear notices on data collection, processing purposes, worker rights and grievance mechanisms. Whether consent from the worker covers everyone else captured on camera — children, relatives, neighbours — remains the unresolved question.

Why does physical AI need home videos?

Physical AI models require real-world data capturing human actions, objects, environments and interactions. Unlike language models, they cannot be trained on scraped internet text. Much of the data must be collected specifically for robotics and embodied AI use cases, which places gig workers with cameras strapped to their bodies at the front line of the supply chain.

Some executives at the meeting raised a sovereignty concern: collecting data in India and selling it to overseas physical AI developers does not, in their view, put the country in a favourable position. Robotics startups went further, arguing India should build domestic physical AI capabilities rather than supply training data to foreign companies. Neither argument has yet translated into policy, and the ministry has not said what rules, if any, it plans to write.

What is Pronto's defence?

Pronto framed its participation as routine engagement. The company said it took part in the MeitY consultation as part of a broader industry discussion on responsible data practices. It said data generated through its customer booking journey involves customer consent and can support downstream AI applications. It added that its existing practices are designed to comply with applicable regulations and that it would continue to engage with the ministry as the policy framework evolves.

Note what the statement does not address: worker consent specifics, retention limits, or whether footage flows to third-party AI developers. The ministry's request for a detailed representation suggests regulators found the answers incomplete.

The commercial stakes are visible in Pronto's cap table and its founder's side project. The Bengaluru-based startup raised $20 million from US-based investor Lachy Groom, co-founder of AI startup Physical Intelligence, months before launching the recording initiative. Founder Anjali Sardana has also set up Provision Labs, a new physical AI venture that works with developers building AI models for robotics and other real-world applications. That arrangement — a services platform collecting video, a founder-adjacent venture serving AI developers — is precisely the kind of structure data-sovereignty hawks at the meeting flagged.

How big is the physical AI money?

Investor interest is measurable. Startups in India's physical AI segment raised roughly $155 million across 31 deals as of July 2026, up from $130 million across 25 deals in 2025, according to an Entrackr report. That is growth of about 19 percent by value year on year — real money, but small next to the billions flowing into US foundation-model developers.

A parallel infrastructure layer is forming. Human Archive builds human sensorimotor datasets for robots. Neo Cambrian has launched an India-focused robotics data factory. Others operating in the segment include iMerit, Human Stryde, Luel, Awign, Build, Centific, Humyn Labs, Aura ML, Build AI, FPV Labs and Northstar. The takeaway: a dozen-plus companies now depend on India-gathered physical data as their core asset, which widens the blast radius of whatever rules MeitY eventually writes.

What happens next?

The development follows earlier discussions among physical AI startups to establish common standards for worker pay, wellbeing, working hours, safety and legal compliance. Those talks produced no binding framework, and self-regulation has now been overtaken by ministerial attention.

Pronto owes the ministry a detailed representation. Until it lands, the industry's default position is split: two of three major platforms refuse in-home recording, one continues, and the legal baseline rests on a data protection act whose worker-specific application has not been tested. For a segment chasing $155 million a year in funding, the cost of getting consent wrong just became a regulatory question rather than a reputational one.