Indian Workers Record Footage to Train AI

News outlets reported that workers in Indian garment factories and private homes have been recorded wearing head-mounted cameras and smartphone rigs to capture first-person or "egocentric" footage of routine manual tasks. AFP published photos and an interview with a Chennai-based worker, Nagireddy Sriramyachandra, who said she is paid 250 rupees (about $2.6) for an hour of video and records household tasks for an AI data company (AFP, June 11, 2026). Dawn published overlapping reporting and named the company Objectways, noting the firm has offices in India and the United States and that it works with Amazon SageMaker and lists Fortune 500 clients (Dawn, June 11, 2026). India Today and multiple social-media posts flagged viral clips of factory lines with workers wearing small cameras on their heads and discussed public fear that such footage could be used to train automation systems (India Today, April 13, 2026).

A May 2026 investigation by Scroll.in (Ayush Tiwari and Raghav Kakkar) identified the specific factory at the centre of the April viral videos as the Gurugram unit of Pearl Global Industries Limited, an apparel manufacturer operating across 10 countries. Two executives from an outside firm arrived in early April with head-mounted devices and distributed them to workers to wear during shifts; workers told Scroll they were not asked for written or verbal consent. The hardware belonged to Egolab.AI, a startup founded in January 2026 by two teenagers -- Raghav Samani, 19, from Sangli, and Varun Pareek, 18, from Ichalkaranji -- that collects "high-quality, labour-sourced egocentric video footage" from factory workers. In March 2026, Egolab was acquired by Build Artificial Intelligence Inc. (Build AI), a Delaware-registered firm led by Edward Xu, 19, and Jonathan Jia, 21. Build AI's CEO claimed $22 million in investment from Abstract, Pear VC, and HF0; HF0 confirmed its investment publicly. Build AI released 100,000 hours of egocentric data on Hugging Face in December 2025. Industry estimates cited in Scroll's reporting put the total robotics labs spending on egocentric data at $1.5 billion to $50 billion over the next two to three years. Egolab's own documents claim clients include Tesla, Boston Dynamics, and Figure AI, though these are vendor-stated claims. Consent issues: Egolab's head of business operations told Scroll the firm "took consent to do it in our own way" but declined to elaborate; Pearl Global did not respond to questions.

First-person or egocentric footage is increasingly valuable for robotics research because it captures hand-eye coordination, object interaction, and environmental context from the actor's viewpoint. Industry research in imitation learning, behavior cloning, and egocentric perception relies on dense, task-specific video plus synchronization metadata (motion sensors, timestamps) to learn policies for manipulation and navigation. Spandan Roy, associate professor at the Robotics Research Centre at IIIT Hyderabad, told Scroll.in that egocentric data offers a cost-effective alternative to recording in three-dimensional simulated environments: "The more data you can feed, more versatility, more variety of processes, scenes, variations that you can embed into the system, the better the robot will function." Practitioners building such datasets typically face challenges with label quality, frame-to-frame alignment, domain shift, and annotation burden. Large-scale consumer or factory recordings lower acquisition costs but increase heterogeneity and noise.

Reporting frames India as a major hub for AI data collection and annotation, supplying large volumes of human-generated training data to global tech companies. Digital labour expert Aditi Surie told AFP that "It's likely that these data collection services will increase," linking the phenomenon to broader demand for real-world robotic data. Astha Kapoor, co-founder of the Aapti Institute in Bengaluru, told Scroll.in that egocentric data falls outside the remit of India's Digital Personal Data Protection Act because it lacks an immediate one-to-one mapping to personal harm: "It is necessary to think about this from the notion of collective data rights because it impacts workers as a collective." The story sits at the intersection of dataset economics, automation risk, and labor rights.

• •Consent and provenance: monitor whether dataset providers publish licenses, chain-of-custody metadata, or client lists making downstream uses auditable.
• •Regulatory responses: follow labor-regulation inquiries, transparency requirements, or guidance on biometric and workplace video collection in India.
• •Client disclosures: note any confirmations from named robotics companies (Tesla, Boston Dynamics, Figure AI) about data sourcing.
• •Market credibility: Build AI's $22M funding claim is CEO-stated; Pear VC and Abstract have not listed Build AI as an investee on their websites as of Scroll.in's publication date.