On 12 November 2026, MeitY Secretary S. Krishnan chaired a meeting with Pronto, Urban Company, Snabbit and robotics companies about footage recorded inside Indian homes to train robots. According to MediaNama's report, which attributes the meeting details to The Economic Times, the ministry asked about data-sharing contracts, retention periods, how much data is collected, whether recordings are linked to workers' identities, how workers consent and are paid, and how the privacy of household members is protected. Pronto was asked to submit a detailed representation on its standard operating procedures.
The strongest case for scrutiny
The case for MeitY's intervention is serious and should be stated fairly. A home is the most sensitive recording environment there is. Footage can capture children, visitors, documents, religious practice and family routines, none of whom signed a services contract. Workers inside the home are in an unequal bargaining position, since refusing a camera may mean losing shifts. And the company's own account has moved. In May 2026, Inc42 reported Pronto's claim that footage is "deleted from our servers in 48 hours" and that it was "not sharing data with any third-party AI labs." By 8 November, per MediaNama, Pronto said it deletes raw footage after the work is done and that, after a 48-hour viewing window, anonymised footage is used for AI training. It did not say who trains on it or where. When the stated purpose shifts from home service to model training, regulators are right to ask.
What the law does and does not settle
The Digital Personal Data Protection Act, 2023 is the relevant framework. PRS India's summary says the Act allows transfer of personal data outside India except to countries the central government restricts by notification, and requires verifiable consent from a legal guardian before processing a child's personal data. MediaNama notes that no country has been restricted so far, that further transfer conditions under the DPDP Rules take effect only in May 2027, and that truly anonymised data falls outside the Act.
That last point is the crux. Pronto's position, reported by Business Today, is that the programme is opt-in, that the customer must affirm it before each booking, and that it covers about 0.1% of customers. Opt-in by the paying customer, however, does not cover the people in the frame who did not opt in: the domestic worker, a spouse, a child, a guest. Blurring faces does not necessarily make home footage anonymous, since voices, layouts, belongings and gait can identify a household. As Apar Gupta of the Internet Freedom Foundation put it in May, a monitoring service in the home and the onward sale of footage to AI labs are two distinct purposes. Each needs its own consent.
The timing gap makes this sharper. The Rules were notified in November 2025, and PTI reported that consent-manager provisions start one year after notification while the broader data-processing obligations arrive after 18 months. Firms are therefore building physical-AI datasets in the interval when the most detailed obligations are not yet enforceable.
Why a ban would be the wrong response
The pro-innovation case is not that anything goes. It is that India has a genuine opportunity in physical AI, and the way to capture it is clear rules, not prohibition. MediaNama reports that robotics startups at the meeting argued India should build its own physical-AI capability rather than only supply data to foreign developers. Some executives also asked MeitY to regulate the sale of such data to overseas AI developers. These are the instincts of an industry that wants a predictable rulebook.
A blanket ban on in-home recording would push training data collection into controlled settings or abroad. Snabbit, for instance, told MediaNama it ran a one-time assessment in a training centre and did not proceed. That is a workable alternative, and Urban Company's founder said in May that it does not record inside homes. The existence of a compliant path shows the market can adapt without the state deciding which robots get built.
A proportionate agenda
MeitY can get most of the benefit without new legislation, using its powers under the existing Act and Rules and guidance it can issue now:
- Separate consent for separate purposes. Service delivery, quality monitoring and AI training should each require distinct, revocable consent, and none should be a condition of getting or doing the job.
- Protect non-consenting people in the frame. Require capture protocols that exclude children and bystanders by default, with a clear rule that footage of minors cannot be used for training without verifiable parental consent.
- Publish retention and recipient facts. If footage or derived data is used for training, firms should disclose in plain language who receives it, in which country, and for how long.
- Worker-specific safeguards. Separate worker consent, disclosed pay for recording, and a bar on linking footage to worker identity for performance or employment decisions.
- A regulator-tested anonymisation standard. Because anonymised data falls outside the Act, "anonymised" should mean something testable, not a face blur.
What to watch
MeitY has so far asked questions and requested a representation. That is the right first step, and it is proportionate. The risk is that questions harden into informal pressure that firms cannot challenge or plan around. The better outcome is a short, published advisory that applies to every company, including those not at the table, before the May 2027 obligations arrive. A clear rule on purpose-specific consent would protect households and workers, and would give India's robotics startups something investors and customers can trust.