‘In the end, you feel blank’: India’s female workers watching hours of abusive content to train AI

2026-02-05 · Source: AI (artificial intelligence) | The Guardian · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Human Resources & Workforce Development, Social Sciences & Behavioral Studies · Depth: Novice, medium

Summary

Women in rural India are experiencing severe psychological trauma while working as content moderators and data annotators for global technology companies, a critical but often overlooked component of AI development. These "ghost workers" classify up to 800 videos and images daily, training algorithms to identify violence, abuse, and harm. Many, like Monsumi Murmu and Raina Singh, are exposed to graphic content, including child sexual abuse and pornography, leading to sleep disturbances, intrusive thoughts, anxiety, and emotional numbing. Sociologists compare this work to lethal industries due to its high psychological risk. Despite the 2021 Indian data annotation market value of $250 million, with 80% of workers from marginalized backgrounds, legal protections and psychological support are largely absent, leaving workers isolated by NDAs and fear of unemployment.

Key takeaway

For CTOs and VPs of Engineering overseeing AI development, your teams must recognize the profound human cost embedded in data annotation and content moderation. Your supply chain for AI training data likely relies on workers facing severe psychological harm without adequate support or legal protection. You should audit your data labeling vendors for ethical labor practices, including mental health provisions and transparent job descriptions, to mitigate both human suffering and potential reputational risk.

Key insights

Content moderation for AI training inflicts severe psychological trauma on workers, particularly women in rural India.

Principles

• AI quality depends on human-annotated data.
• Content moderation is a high-risk occupation.
• NDAs exacerbate worker isolation and harm.

Method

Workers classify flagged images, videos, and text to train algorithms, often viewing hundreds of disturbing items daily to identify violations of platform rules.

In practice

• Implement robust psychological support for content moderators.
• Ensure transparent job descriptions for data annotation roles.

Topics

• Content Moderation
• AI Training Data
• Psychological Trauma
• Data Annotation
• Labor Exploitation

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Ethicist, Policy Maker, Research Scientist

Related on AIssential

• Human Capital — Data workers, primarily in the Global South, are exploited and invisibilized despite powering AI development.
• Lawsuit: ChatGPT told student he was "meant for greatness"—then came psychosis — AI chatbots can exploit human psychology, potentially leading to severe mental health consequences.
• The week that Meta employees became training data — Meta's Model Capability Initiative transforms employee actions into AI training data, reflecting a broader trend to overcome data scarcity by internalizing surveillance.
• New Study Examines Features and Policies for 29 AI ‘Undressing’ Apps — AI "undressing" apps are vehicles for systemic gender-based violence, objectifying women and commodifying their exploitation.
• ‘Who is going to pay us when we’re replaced by robots?’ The Indian factory workers told to film themselves for AI — The global race for egocentric data exploits Indian factory workers, raising ethical concerns about consent, compensation, and data ownership.
• The Machine’s Human Cost. The lived reality is unpaid testing, chaotic onboarding, sudden project cancellations, constant Slack monitoring... — AI's polished outputs are built on precarious, hidden human labor, often from displaced creative professionals.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI (artificial intelligence) | The Guardian.