Freedom, for most Indian women, has never been about freedom to earn, benchmark for the freedom has been them whether a woman is "allowed to go alone" to the local market, a health facility, or outside her village. Geography of this freedom is not random, it is dense precisely in the Hindi-speaking belt, where states like Bihar, Uttar Pradesh, Rajasthan, and Jharkhand account for less than 18 percent of working women getting paid for their labour, while Telangana and the southern and north-eastern states performed comparatively better.
This phenomenon is not a rural problem alone, this conditioning for Indian women, began at home long before any AI company arrived. It was the door she couldn't walk through alone. The money she helped earn but didn't get to spend. Decisions about her own health, her own life, made for her by a husband or a mother-in-law. Even the work she found outside the home offered little relief. Limit to that freedom was drawn in daily cooking, the cleaning, and raising of children never being counted as work at all. No skill. No wage woman was expected to do, for free. Women spend about 5 hours a day (289 minutes on unpaid domestic services plus 137 minutes on caregiving) on unpaid household work, versus roughly 2.5 hours for men (88 and 75 minutes respectively.
AI companies did not create this architecture of unfreedom they simply arrived at its door, camera in hand. Long before any machine could learn to fold laundry, the folding itself was already hers to do, and had been for centuries. What these companies brought was not a new demand of her labour, but a new way of guarding her, camera fitted to the head, the chest, a pair of glasses, like a live frame placed directly over an architecture patriarchy had already built. It doesn't ask her to do anything she wasn't already doing. It only asks her to keep doing it in view capturing exactly what her eyes would see as her hands fold the cloth, stir the pot, assemble the pieces she has assembled a thousand times before, unwatched and unpaid but this time they arrived with something patriarchy never offered an immediate cash income ₹250 or $1 to $4 per hour, a sense of identity, means of employment and source of income.
Meanwhile these companies did not prepare any procurement document, recruitment ad, training manual, or internal company presentation or a policy to target these women Which can be claimed with evidence. Tasks themselves (folding towels, arranging objects, slicing fruit, making garlands) were decidedly coded as domestic/feminine labour in Indian social context. Companies did not need to formally “map” this skill distribution because it is already extremely well documented at the population level by India’s own statistical apparatus. For example, India’s Time Use Survey shows women spend roughly 299–315 minutes a day on unpaid activities, compared with 56- 97 minutes for men, depending on the year and definition.
Companies like Pronto, Human Archive, Humyn Labs, EgoData/EgoLab, Objectways, Neo Cambrian AI, Awign, iMerit, XP Robotics, DataScanAI, Nxted, Qanat Consulting Services, MoVo, HumanStryde, Field Motion, Market Xcel, Nexdata, Daidai Labs, and Egocentric Network recruit through Self-Help Groups (SHGs), the National Rural Livelihood Mission (NRLM/DAY-NRLM), Anganwadi networks, Panchayat structures, or microfinance institutions specifically for embodied-AI data work.
Companies market this work in the language of liberation: "flexible," "empowering," income earned "without needing to travel or leave the home." It is framed as something that fits neatly alongside existing duties, not something that require do additional work that they are already doing.
For some women, this advertisement isn’t just marketing. Hope for autonomy is real. An independent income, covering her own expenses for the first time, or finding a way out of a financially controlling or abusive situation.
a woman’s domestic labour, mango-slicing, an ordinary task that she would do unpaid regardless, becomes, domestic task, a recorded performance, a dataset asset and a component of a commercial AI product, the moment a camera stands on her head. Who else will pay 250 rupees an hour just for doing housework? This work provides something that looks like option for work from home, record from personal kitchen, no commute, no office, no need to ask permission to leave house for work.
But this freedom quietly takes a form of restriction, which keeps women out of the formal workforce, Following the script of approving women for work. Normalizing women must earn as long as its home based and perform household duties, where family benefits too, aligning with women from respected family don’t work outside.
Pronto, the Bengaluru-based on-demand house-help startup was linked the company's operations to “physical AI" and robotics training data. It had launched a pilot in which domestic workers wore head-mounted, outward-facing cameras while performing tasks like cleaning, laundry, and meal preparation inside customers' homes, with the resulting footage processed to train physical AI systems and robots to replicate human hand movements and working processes. It was accused of customer-facing trust issues and safety feature and anonymized, audio-free footage collection.
While in Tamil nandu Karur textile and Qanat Consulting Services are startups in India that recruit people to record first-person or egocentric videos. Many of them record data, check quality parameters, and perform data annotation on behalf of clients in the U.S. and China. Neocambrian AI, a Noida based startup collects robotics data of more than 100 factories where workers record themselves performing tasks, which is creating datasets which are “useful for solving handiness,” and teaching robots “how to manipulate objects.
Surveillance inside the home where cameras are, by construction, recording inside domestic space the privacy is becoming a stake. Some workers were not even told their footage would train AI models; some were told a false purpose like hours-monitoring.
When a home becomes the dataset, privacy is no longer limited to the individual worker, it extends to the entire domestic environment. Even when faces are blurred or absent, workers may remain identifiable through linked records, while the recording can potentially capture family members, children, possessions, and intimate details of the home.
This is a documented consent failure, Workers are majorly paid for the act of being filmed, not for the resale value of the resulting intellectual property. Minimum wage, occupational safety, and maternity-benefit provisions does not even apply to this work, because Under the Code on Social Security, 2020, "gig worker" is defined to cover platform-based work, and definition provided for only ride-hailing and delivery, not for a worker whose primary output is a resellable dataset.
There is no labour category for "data-recording work," no minimum-wage benchmark for an hour of egocentric footage, no statutory line between "monitoring an employee" and "manufacturing a commercial AI asset". which is why companies like Egolab, snabit and Pronto describe the same act of filming a worker as, respectively, a factory efficiency measure and a customer-service feature. Section 7 of DPDP Rules, 2025, It only applies to "personal data breach." Neither Pronto nor Egolab has (yet) had a reported breach and legally routine collection is consented till no personal data breach is reported, So Rule 7 does not apply on it.
Feminist technology scholars point out that outsourcing domestic work to mechanical surrogates does not automatically dismantle household gender inequality. Instead, it often mechanizes traditional stereotypes, positioning the robot as an electronic housewife while maintaining unequal structural power dynamics within the home. The expensive embodied AI products eventually built from this localized data will be marketed to wealthy consumers, doing little to alleviate the material realities of working-class or marginalized women who shoulder the heaviest burdens of unpaid household labour.
If we particularly talk about the women, then comparison between freedom for women before and after being employed by ai companies reveals a paradox of “freedom through unfreedom.” Patriarchal domestic labour is unpaid, socially invisible and largely unmeasured, while embodied-AI labour makes women’s work paid, visible and precisely measurable. Yet this recognition comes through intensified surveillance and standardisation movements are recorded, restrictions are imposed, and workers have little control over how their labour is used.
Commercialization of domestic labour allow patriarchal norms to adapt rather than disappear. Patriarchy, which historically restricted women’s mobility by defining the home as their proper sphere, can potentially adapt to AI-enabled home-based work: “You can earn, but you do not need to leave the house.” In this sense, embodied-AI labour may convert the home from a site of unpaid reproductive labour into a site of paid productive labour without fundamentally challenging women’s spatial confinement.
India’s embodied-AI data labour is part of a wider Global South pattern in which AI companies outsource low-paid, tightly controlled data work to workers with limited bargaining power. Across the Philippines, Kenya and Latin America, similar patterns of low wages, platform dependence and labour precarity appear, leading to “triple burden” for women of unpaid care work, economic insecurity and platform volatility. In this quest of high racing AI race India has positioned itself as a global middleman for the creation, processing and annotation of AI data. But Question why India, because it has no labour classification for "data-recording gig work," so no regulation governs it.
