On 22nd May 2026, a media report disclosed that Pronto, a platform designed to connect trained household service professionals with those in need of help with household chores, is inclined to build a data layer through camera-facilitated surveillance inside users’ homes. According to the report, Pronto’s service professionals were wearing outward-facing cameras for recording tasks such as washing dishes, folding laundry, cleaning homes and meal preparation. The purpose behind these practices is to train physical AI and mitigate the data abscess for physical AI development. Even though the Ministry of Electronics and Information Technology is reportedly looking into the matter, such practices pose serious concerns for digital privacy, labour rights and the future of work in a world that is slowly beginning to explore pathways to substitute human effort with data-driven automation.
In a blog, Pronto mentioned how physical AI labs have struggled to advance due to the lack of data, especially data that includes first-person video-based surveillance of their own trained professionals undertaking tasks requested by consumers through their platforms. At the same time, they assured that this feature would not be enabled for a majority of users and is being tested through a pilot program covering merely 0.1 per cent of its user base. This reassurance, however, seems more like a half-measure than a substantive statement. The company’s true intentions seem unclear as its statements citing the dearth of data powering physical AI contradicts how it wishes to leverage gig-worker facilitated camera surveillance. Why cite lack of quality data as a problem statement and participation of gig workers in the AI economy as an incentive when it’s never going to be rolled out to the majority of consumers?
Further, research suggests that face-blurring technologies are fundamentally inadequate as a privacy-preserving measure. Citing that work has been undertaken to ensure ‘above and beyond’ compliance with the law, Pronto has stated that it will blur facial biometrics automatically and no personally identifiable information will be ever uploaded or shared. It is also unclear whether the company’s anonymisation measures are sufficiently robust to prevent the re-identification of individuals or households captured in the data. Even where facial biometrics are blurred, footage recorded within private homes may still contain unique features such as room layouts, personal belongings, family photographs, voices, locations, or other contextual information that could reveal the identity of individuals. In highly personal spaces such as homes, true anonymity is difficult to guarantee, this raises concerns about whether such datasets can ever be fully de-identified and distanced from a unique digital footprint before being used for AI training purposes.
It is claimed that the footage would also be deleted within 48 hours. It is also claimed that other than the consumer themselves, no one will have access to the footage. Again, this contradicts their earlier statements to build an effective data layer. If all footage will be deleted within 48 hours and if only the user has access, there is no way in which this exercise will enable data-driven advancements in physical AI use cases.
Another aspect of privacy that comes under scrutiny relates to the opt-in mechanism. While a user may consent to the recording of household service professionals for purposes such as safety or theft prevention, that consent cannot automatically be stretched to cover the use of footage for training AI models, where intimate details of their homes may become part of valuable training datasets. The collection of data for one purpose does not necessarily justify its use for an entirely different purpose, particularly where such use involves the development of commercial AI systems. Additionally, while the opt-out feature is reportedly available to users, it remains unclear whether similar choices are available to household service professionals. It is also unclear whether Pronto has taken adequate steps to inform workers about why their activities are being recorded, how the resulting data may be used, and whether they can meaningfully withdraw their consent at any stage. In the context of gig work, consent is rarely exercised on an equal footing. Household service professionals may feel compelled to participate in such schemes to retain access to work opportunities, making any purported consent less than fully informed or freely given.
The Digital Personal Data Protection Act, 2023 is also unlikely to provide immediate relief (especially in cases where the face blurring technology fails and the data is leaked) if users are consenting to data collection practices that will record, detect and blur their facial biometric data. Substantively, consumers as Data Principals will have limited recourse, especially considering the fact that grievance redressal processes under the DPDP framework allow a 3 month response time to Data Fiduciaries when a grievance is raised. There are other aspects of the grievance redressal process that also raise substantive concerns on the effectiveness of this process.
Then there are broader questions on its significance on gig workers’ rights. Such pilots not only deprive gig workers’ of any agency in becoming conduits for enabling camera-based surveillance but also raise serious concerns on the future of work. There are numerous instances where companies have begun meticulous surveillance tactics to collect data on its own employees while they are performing their jobs to fulfill their commercial incentives. If Pronto’s statement is true, then there also remains no compelling reason to test this in a pilot form altogether. Again, there is a stark misalignment in their reassurances, motivations and ambitions. Why embrace physical AI and allow their own professionals to participate in the AI economy if this is never going to be rolled out to the large majority of the Indian consumer base?
Another important concern is that if this data is ultimately used to train physical AI systems, then questions arise as to who owns the value of the labour-generated data itself. Every recorded movement, gesture, task sequence, and interaction inside a home contributes to the development of automation technologies that may eventually reduce the need for human labour itself. Yet, workers appear to receive neither compensation, recognition, nor bargaining power in exchange for producing this data. This is also particularly concerning because, unlike other companies that collect first-person or “egocentric” video data through dedicated programs where participants knowingly record themselves for the express purpose of AI training and are compensated for their contributions, Pronto appears to be leveraging ordinary service interactions between consumers and household service professionals to generate training data. In such cases, workers sign up to provide household services which is their primary job for which they are paid, not to become data annotators or contributors to AI development but continue to do so without any specific payment for it. Similarly, consumers engage the platform to access household services, not to have their homes incorporated into datasets used for commercial AI development. This blurs the distinction between service delivery and data extraction, and the question about whether workers and consumers are adequately informed about the downstream uses of the data being collected remains. It also raises critical concerns about whether gig workers are unknowingly becoming unpaid contributors to the AI supply chain, where their physical labour is transformed into proprietary datasets that primarily benefit platforms and technology developers rather than the workers generating them. In fact, they are unknowing contributors to a system that is designed to replace them or reduce their job security.
In the recent past, Pronto was subjected to public scrutiny when a customer reported an incident where a grievously injured Pronto worker was compelled to finish the work despite her physical condition. Ironically, Pronto’s statements raise more questions on gig workers’ participation in the AI economy. Such incidents are not merely isolated examples but are also indicative of the systemic injustices of being a gig worker in the platform economy. In a survey performed by Communication Workers of America, it was found that the majority (52%) of surveyed workers believed that they are training AI to replace other workers’ jobs, and 36% believed they are training AI to replace their own jobs. An article in the MIT Technology Review provided case studies of people in India and Nigeria strapping mobile phones on their heads to record themselves doing their work. Moreover, in the recent past, there have been two separate instances reported in India itself where factory workers are wearing head-mounted cameras to record themselves during their shifts. Whether in India or globally, the number of companies exploring ways to automate human effort are gradually increasing. This results in either the company leveraging its in-house human workforce to train and build AI-driven robotics or engaging independent startups to collect such video-based data and selling it to companies who can further leverage it for automation. Either way, the increasing number of such instances pose some real and uncomfortable questions on humanity’s role in shaping the future of work and how it will interact with emerging technologies.
